Formless forms and paper-based user interfaces

0001The present invention relates to user interfaces for computers and information processing systems and networks, and more particularly to paper-based user interfaces.

0002As is well known, the user interface is the gateway through which a human user communicates with a computer, computer network, or other system for processing digital information. For example, a desktop personal computer's user interface can include a keyboard, a mouse or other pointing device, and a display screen, coupled with appropriate software such as a command-line interface or a window-and-icon or other graphical user interface (GUI).

0003As is also well known, it is now commonplace for computers, printers, optical scanners. and other devices, such as multifunction devices (standalone devices that offer a combination of printing, copying, scanning, and facsimile functions), to be connected to and through local-area and wide-area computer networks. For example, this text is being written on a personal computer that is connected through a local-area network,to a server computer, which controls the hard disk drive where the text will be stored, and also to several printers, on which the text can be printed, as well as to an Internet gateway server, which connects the personal computer with the Internet. The same local-area network can further accomodate a scanner or a multifunction device having a scanning capability, so that paper documents or portions thereof can be scanned and converted (as by optical character recognition software) to useable form, and then sent to the personal computer for editing or other processing.

0004The widespread availability of optical scanners, facsimile (fax) machines, multifunction devices, and other devices and subsystems by which computers and computer networks can "read" paper documents gives rise to the concept of a paper-based user interface. A paper-based user interface allows the user of a computer, computer network, or other digital information processing system to communicate with the system simply by making a mark or marks on a paper document or documents and then scanning the document thus marked into the system via a scanner, fax machine, multifunction device, or the like. The system can communicate back with the user by printing another document.

0005Paper-based user interfaces are known, if not yet commonplace. In particular, an exemplary paper-based user interface is known that allows a personal computer to receive and respond to user commands sent remotely from a distant facscimile machine. The user is provided with a blank form that has been printed on paper. The user fills out the form by making marks (as with a pen or pencil) in specific designated fields on the form. For example, the user can enter a check-mark or an X in a blank box on the form. Then the user sends a facsimile of the form thus marked to the personal computer. The personal computer receives the facsimile transmission, for example, through a modem, and so is provided with a facsimile copy of the user's marked-up form. The personal computer can then interpret the marks on the form by running image processing and other special software for this purpose, and can respond to the marks accordingly by carrying out commands that correspond to the particular boxes the user has marked. For example, the personal computer can retrieve an electronically stored document that the user has requested and send this document to the user by return facsimile. Thus the user and the computer "fax" each other in much the same way that two human beings can communicate by sending facscimiles back and forth to one another.

0006A paper-based user interface can serve as a complement or substitute for the more conventional keyboard-mouse-display type of user interface mentioned earlier. A paper-based user interface is particularly appealing when the user interacts with a computer network directly through a multifunction device, without recourse to a personal computer or workstation. In this situation, the user can initiate a number of functions, such as document copying, facsimile, electronic mail, document storage, and search using a simple paper form as an interface. The multifunction device "reads" what is on the form and responds accordingly, possibly with help from the network.

0007Paper-based user interfaces typically require that forms be created in advance, either by the user with a form editor or automatically by computer, so that the receiving computer can readily determine whether and where a given form has been marked by the user. For example, suppose that a particular form contains a set of blank boxes in which the user can enter check-marks or Xs to indicate certain requests. The user selects the form, checks some of the boxes, scans the form into the system to produce a digital image, and transmits this image (more precisely, transmits data representing the image) to a computer. Upon receiving the transmitted image of the user's marked-up form, the computer compares the image with a stored representation of the unmarked form. Based on the results of the comparison, the computer can tell what the user has requested and take any action appropriate in response. In order to make the comparison, however, the computer must first have the information necessary to interpret the form, such as information about where the blank boxes are located on the form, how big the boxes are, and what each box means, that is, how the computer should respond when certain boxes are marked. This information can be provided to the computer either in advance of the user's transmission, or concurrently with or as part of the user's transmission. For example, the computer can be given access to a set of stored digital representations each indicating the layout or appearance of one of a set of forms, and the user can transmit along with the marked-up form image an identification number that uniquely corresponds to the particular type of form being used. As another example, specially coded information, such as a pattern of data glyphs or a bar code, can be included in the form itself to indicate the layout of the blank fields in the form. The computer can be programmed in this case to seek the coded information at a predesignated location within the received image, and to use the coded information together with additional (stored or preprogrammed) information to identify what kind of form has been sent and to determine what is to be done in response to the boxes checked by the user.

0008Known paper-based user interfaces are greatly limited in the possible appearance and layout of forms they support. More particularly, the forms have been limited to whatever can be readily constructed with a forms editor or automatic forms generation program, and the active elements of the forms have been restricted to very simple graphical elements such as check boxes, open circles or ovals, rectangular blank spaces, blank lines, and the like. An example of a form used in a paper-based user interface in the prior art is seen in FIG. 1, which is a Universal Send Form from the PaperWorks software product formerly available from Xerox Corporation (Stamford, CT). The form includes check boxes such as check boxes <b>1</b> and blank rectangles such as rectangles <b>2.</b> as well as data glyphs <b>3</b> that are used by the computer to help it interpret the form

0009The range of application of paper-based user interfaces has therefore been limited. If the potential of these interfaces is to be fully realized, a more flexible and powerful way to create forms is needed.

0010The present invention offers a new approach to computer-readable forms, wherein arbitrary documents can be used as forms. This approach can be called <i>formless forms.</i> According to the embodiments of invention, a stored digital document of almost any kind can be used as a form for a paper-based user interface, even if the document was not originally designed or intended for use as a form. For example, in certain embodiments, a document containing an arbitrary combination of text, graphics, and bitmap images can be used, and any area of the document can become an active region or field of the form. Formless forms offer new flexibility when applied in existing paper-based user interfaces, and also can provide new applications for paper-based user interfaces. One such application is in a paper-based World Wide Web user interface, here called PaperWeb.

0011In one aspect of the invention there is provided a method carried out in a data processing system, comprising: providing a first document image comprising digital image data including a first plurality of pixels, the first document image representing an instance of a reference document to which instance a mark has been added, the reference document having a plurality of elements; providing a second document image comprising digital image data including a second plurality of pixels, the second document image being selected from among a plurality of document images and representing the reference document without the mark; automatically extracting from the first document image a set of pixels representing the mark by performing a reference-based mark extraction technique wherein the second document image serves as a reference image and wherein substantially the entirety of the first document image is compared with substantially the entirety of the second document image; providing information about a set of active elements of the reference document, each active element being one element among the plurality of elements of the reference document, the reference document having at least one such active element, each active element being associated with at least one action; interpreting the extracted set of pixels representing the mark by determining whether any of the active elements of the reference document is indicated by the mark; and if an active element is indicated by the mark, facilitating the action with which such active element is associated.

0012Preferably, the steps of registering the first and second images and computing a robust difference of the registered images are performed piecewise on local subregions of the document images.

0013Preferably, the interpreting step comprises determining whether the mark is proximate to any active element.

0014In another aspect of the invention there is provided a method according to claim 6 of the appended claims.

0015Preferably, the step of computing a robust difference comprises: determining a collection of pixels common to both the first and second images by matching pixels of the first image with pixels of the second image according to a matching criterion; and eliminating as between the first and second images the pixels of the collection thus determined, thereby determining a set of discrepancy pixels including a set of pixels representing the mark.

0016The invention further provides a method according to claim 7 of the appended claims.

0017The step of converting the second document instance may comprise scanning a hardcopy instance of the second document with a digital scanning device.

0018The step of converting the second document instance may comprise rendering a symbolic description of the second document into a bitmap form.

0019The step of converting the second document instance may comprise scanning a collection of hardcopy document instances including the second document instance with at least one digital scanning device, thereby producing a collection of scanned document instances, and storing the scanned document instances in a database accessible to the processor; and the providing step may comprise retrieving the second document image from the database by generating an image-based index from the first document image without recognition of any symbolic content of the first document image and selecting a second document image from among the scanned document images in the database.

0020Preferably, the annotating step is performed prior to the providing step and comprises storing the set of annotations and wherein the providing step comprises contemporaneously retrieving the second document image and the set of annotations.

0021The invention further provides a method according to claim 8 of the appended claims.

0022Preferably, the first providing step comprises providing the processor with a first document image wherein the selected element is an element other than a form element selected from the group consisting of: a substantially blank region having a perimeter delimited at least in part by a perimetral boundary and within which blank region a substantial portion of the mark is disposed; a substantially blank region substantially underscored by a baseline proximate to which baseline a substantial portion of the mark is disposed; a textual word located in an array of textual words set apart from any surrounding text, the mark being proximate to the textual word; and a graphical symbol located in an array of graphical symbols set apart from any nearby nonblank elements, the mark being proximate to the graphical symbol.

0023Preferably, at least one action with which each active element is associated is an action of following a specified hypertext link; and    the facilitating step, if performed, comprises initiating a traversal of the specified hypertext link.

0024The invention further provides a programmable data processing system when suitably programmed for carrying out the method of any of claims 1 to 8 of the appended claims, or according to any of the particular embodiments described herein.

0025The invention will be better understood with reference to the drawings and detailed description below. In the drawings, in which like reference numerals indicate like components: <ul id="ul0001" list-style="none" compact="compact"><li>FIG. 1 (PRIOR ART) is an example of a form used in a known paper-based user interface;</li><li>FIG. 2 schematically depicts the components of a system suitable to an embodiment of the present invention;</li><li>FIG. 3 schematically depicts the components of a computer from the system of FIG. 2;</li><li>FIG. 4 flowcharts the steps of a method for Formless Forms in one embodiment of the invention;</li><li>FIG. 5 schematically depicts the information flows in the method of FIG. 4;</li><li>FIG. 6 is a flowchart showing in more detail the indexing step of FIG. 4;</li><li>FIG. 7 is a flowchart showing in more detail the mark extraction step of FIG. 4;</li><li>FIGS. 8A-8B schematically depict how an image can be divided into regions for purposes of registration and differencing;</li><li>FIG. 9 flowcharts a method for assigning active elements to a Formless Form;</li><li>FIGS. 10-12 are a series of views that illustrate mark extraction for a test image;</li><li>FIGS. 13-19 are a series of views showing an illustrative example of processing a Formless Form according to the steps of the method of FIG. 4;</li><li>FIG. 20 is an example of a marked-up World Wide Web page printout that can be used as input to Paper Web;</li><li>FIG. 21 is a series of views that depicts the inputs and outputs of an exemplary interchange between a user and a computer running a PaperWeb browser;</li><li>FIG. 22 flowcharts a method for Paper Web in one embodiment; and</li><li>FIG. 23 flowcharts a method for assigning active links in Paper Web.</li></ul>

<i>Overview</i>

0026A paper-based user interface (PUI) provides a user interface to a computer, computer network, or other computational or information processing system in which a paper (or other hardcopy) document instance is marked up by a user, scanned into the system, and interpreted by the system in order to control some computational process Formless Forms makes it possible to use arbitrary documents, rather than specially designed forms, as the basis for a PUI. Paper Web applies Formless Forms PUI technology to provide a new kind of browser for the World Wide Web and other hypertext databases.

0027According to the invention, an ordinary paper document can become, in effect, a "form"; that is, it can be made recognizable by a PUI and used to control arbitrary computational processes. Associative links can be established between graphical or other compositional elements of a document and computational processes to be controlled. Any textual, graphic, photographic, or other compositional element of the document can be made computationally active (that is, can be treated as a cue to trigger an arbitrary computation or computations). For example, with Formless Forms, a character, a word, or a graphical element can be marked by the user and the computer will respond by carrying out an action or actions associated with the marked item. Notably, the computer need not recognize the marked character, word, or graphical element as such in order to carry out the actions associated with the marked element. That is, the computer need not perform optical character recognition (OCR), word recognition, linguistic analysis, pattern recognition, or other symbolic processing. Instead, the computer can perform its processing in the image domain, using image processing techniques to distinguish (extract) the marks from the original unmarked document and thus to determine where the user has marked the document and which elements of the document are indicated. The computer can then readily determine whether and what computational processes are associated with the indicated elements, and respond accordingly.

0028In short, with Formless Forms, a computer can respond to a mark made on a document that was never intended for use as a form in the same way as if the user had marked a check box on a conventional form. Moreover, the computer can do this without symbolic reasoning and without recognizing the semantic content of the document.

<i>System Components</i>

0029FIG. 2 schematically depicts an example of a computational system <b>10</b> suitable to an embodiment of the system and method of the invention. System <b>10</b> includes a fax machine <b>20</b>, a "smart" multifunction device <b>30</b> (that is, a multifunction device incorporating a processor (CPU) and memory), a personal or office computer <b>100</b>, one or more local server computers <b>40</b>, and one or more World Wide Web server computers <b>50.</b> These are connected by various communications pathways including telephone connections <b>11,</b> a local area network <b>41,</b> and the Internet <b>51.</b> Computer <b>100</b> includes a modem <b>108</b> and optionally a CD-ROM mass storage device <b>109,</b> and has attached peripherals including an optical scanner <b>103</b> and a printer <b>104</b>.

0030Persons of skill in the art will appreciate that the design of system <b>10</b> is intended to be illustrative, not restrictive. In particular, it will be appreciated that a wide variety of computational, communications, and information and document processing devices can be used in place or or in addition to the devices <b>20, 30, 40, 50</b>, and <b>100</b> shown in system <b>10</b>. Indeed, connections through the Internet <b>51</b> generally involve packet switching by intermediate router computers (not shown), and computer <b>100</b> is likely to access any number of Web servers <b>50</b> during a typical Web browsing session. Also, the devices of system <b>10</b> can be connected in different ways. For example, printer <b>104</b> is shown as being an attached peripheral of computer <b>100</b>, but it could also be a networked printer, accessed via local area network <b>41</b> through a print server that is one of the local servers <b>40.</b>

0031The various communication pathways <b>11, 41, 51</b> in system <b>10</b> allow the devices <b>20, 30, 40, 50, 100</b> to communicate with one another. Telephone connections <b>11</b> allow fax machine <b>20</b> to communicate with multifunction device <b>30,</b> and also with computer <b>100</b> by way of modem <b>108.</b> Local area network <b>41</b> allows computer <b>100</b> to communicate with local server(s) <b>40.</b> The Internet <b>51</b> allows multifunction device <b>30</b> and computer <b>100</b> to communicate with Web server(s) <b>50.</b>

0032A wide variety of possibilities exists for the relative physical locations of the devices in system <b>10</b>. For example, fax machine <b>20</b> and multifunction device <b>30</b> can be in the same building as each other or around the globe from one another, and either or both can be in the same building as computer <b>100</b> or around the globe from computer <b>100</b>. Web server(s) <b>50</b> can likewise be at local (so-called "Intranet") or remote sites with respect to computer <b>100</b> and multifunction device <b>30.</b> The distance between computer <b>100</b> and local server(s) <b>40</b>, of course, is limited by the technology of local area network <b>41.</b>

0033A user or users can access system <b>10</b> at various points and in various ways. For example, a user can provide inputs to and receive outputs from system <b>10</b> through fax machine <b>20,</b> through multifunction device <b>30,</b> or through the scanner <b>103</b> and printer <b>104</b> of computer <b>100</b>. In particular, a user who is near fax machine <b>20</b> can send a fax from fax machine <b>20</b> to computer <b>100.</b> and computer <b>100</b> (assumed here to be suitably programmed with Formless Forms PUI software) can automatically send a fax back to the user at fax machine <b>20</b>. Similarly, the user can send a fax from fax machine <b>20</b> to multifunction device <b>30</b> and multifunction device <b>30</b> (likewise assumed to be suitably programmed) can automatically send a fax back to the user at fax machine <b>20</b>. A user who is near computer <b>100</b> can interact with computer <b>100</b> through its PUI in conjunction with scanner <b>103</b> and printer <b>104</b>. A user who is near multifunction device <b>30</b> can interact with multifunction device <b>30</b> through its scanning and printing capabilities, thereby using multifunction <b>30</b> as a kind of personal computer, a computer having a user interface that is primarily or even exclusively paper-based. Finally, the user can interact with Web server(s) <b>50</b> by browsing the Web. This can be done directly from computer <b>100</b> or multifunction device <b>30</b>, or indirectly from fax machine <b>20</b> by way of either computer <b>100</b> or multifunction device <b>30.</b>

0034FIG. 3 schematically depicts the components of computer <b>100</b> of the system of FIG. 2. Computer <b>100</b> is a personal or office computer that can be, for example, a workstation, personal computer, or other single-user or multi-user computer system. For purposes of exposition, computer <b>100</b> can be conveniently divided into hardware components <b>101</b> and software components <b>102</b>; however, persons of skill in the art will appreciate that this division is conceptual and somewhat arbitrary, and that the line between hardware and software is not a hard and fast one Further, computer <b>100</b> is shown as including peripheral components, viz, scanner <b>103</b> and printer <b>104</b>; again, it will be appreciated that the line between host computer and attached peripheral is not a hard and fast one, and that in particular, components that are considered peripherals of some computers are considered integral parts of other computers.

0035Hardware components <b>101</b> include a processor (CPU) <b>105,</b> memory <b>106</b>, persistent storage <b>107,</b> modem <b>108,</b> optional mass storage <b>109,</b> and network interface <b>110</b>. These components are well known and, accordingly, will be explained only briefly.

0036Processor <b>105</b> can be, for example, a microprocessor or a collection of microprocessors configured for multiprocessing. It will be appreciated that the role of computer <b>100</b> can be taken in some embodiments by multiple computers acting together (distributed computation); in such embodiments, the functionality of computer <b>100</b> in system <b>10</b> is taken on by the combination of these computers, and the processing capabilities of processor <b>105</b> are provided by the combined processors of the multiple computers.

0037Memory <b>106</b> can include read-only memory (ROM), random-access memory (RAM), virtual memory, or other memory technologies, singly or in combination. Persistent storage <b>107</b> can include, for example, a magnetic hard disk, a floppy disk, or other persistent read-write data storage technologies, singly or in combination.

0038Optional mass storage <b>109</b> provides additional (e.g., archival) storage and can be a CD-ROM (compact disc read-only memory) or other large-capacity storage technology. In some embodiments, mass storage is provided by one of the local servers <b>40</b> rather than as part of computer <b>100</b>, for example, a file server or database server. In some embodiments, mass storage as such is omitted, as where persistent storage <b>107</b> is of sufficient capacity to meet the needs of computer <b>100</b> that would otherwise be served by mass storage.

0039Network interface <b>110</b> provides computer <b>100</b> and, more specifically, processor <b>105</b> with the ability to communicate via local-area network <b>41</b> and (either directly, or indirectly via one of local servers <b>40</b>) with the Internet <b>51.</b>

0040Notably absent from hardware components <b>101</b> are a keyboard, mouse or other pointing device, and display screen. Such components, while typically part of most computers, are not necessary to the invention. This is because such components are adapted for providing (for example) a graphical user interface, whereas here the focus is on a paper-based user interface.

0041Software components <b>102</b> include a multitasking operating system <b>150</b> and a set of tasks under control of operating system <b>150,</b> such as applications programs <b>160, 161,</b> and <b>162.</b> Operating system <b>150</b> also allows processor <b>105</b> to control various devices such as persistent storage <b>107</b>, modem <b>108</b>, mass storage <b>109</b>, and network interface <b>110</b>, as well as peripherals <b>103.</b><b>104</b>. Processor <b>105</b> executes the software of operating system <b>150</b> and its tasks in conjunction with memory <b>106</b> and other components of computer system <b>100.</b>

0042Software components <b>102</b> provide computer <b>100</b> with Formless Forms PUI capability. This capability can be divided up among operating system <b>150</b> and applications programs <b>160, 161. 162</b> as may be appropriate for the particular configuration of system <b>10</b> and the particular application of computer <b>100</b> in system <b>10</b>. For example, operating system <b>150</b> can incorporate fax server and Internet server software, and the Formless Forms and Paper Web methods described below can be incorporated in an applications program or programs. It will be appreciated that there are any number of possibilities in this regard.

<i>Formless Forms Method and Examples</i>

0043FIG. 4 is a high-level flowchart for the method of Formless Forms in one embodiment. These steps are carried out using appropriate components of system <b>10</b> and. more particularly, computer <b>100</b> under control of processor <b>105.</b>

0044To begin with (step <b>410</b>), an image (more precisely, a set of digital data representing an image) of a marked document instance is obtained in computer <b>100</b> and is stored in memory <b>106</b> for use by processor <b>105</b>. The image can be a scanned, faxed, or other pixel (raster) image. The image can be scanned in from scanner <b>103</b> directly to computer <b>100</b>, or can be input to system <b>10</b> through fax machine <b>20</b> or multifunction device <b>30</b> and communicated by the appropriate pathways to computer <b>100</b>. Alternatively, the image can be an image that previously has been scanned and stored in mass storage <b>109,</b> and retrieved therefrom by computer <b>100</b>. It is assumed in this embodiment that this image and all document images are black-and-white binary images, although the image can be gray-scale or color in other embodiments. Typically in this embodiment, a document image is an image of a single page or a portion thereof, although a document image can encompass multiple pages in some embodiments.

0045The image is an image of a marked document instance. That is, it is an image of a particular hardcopy instance of a document (e.g., a printed, faxed, or reprographically produced paper copy, or a hardcopy on a medium other than paper) that the user has in her or his possession and on which instance the user has made a mark (e.g., a graphical or written annotation or addition to the basic document) with a pen, pencil, typewriter, rubber stamp, etc.

0046The image is said to be an image of a "document instance," rather than simply of a document, to highlight the fact that it is an image of a particular (and usually imperfect) copy, printout, or other rendering of a known document for which a (presumably good) original or other reference version has been stored in a database or other collection of documents present somewhere in system <b>10.</b> The collection of documents can be stored, for example, in mass storage <b>109</b>, or on a local file or database server <b>40,</b> or on a Web server <b>50.</b>

0047Once the image of the marked document instance has been stored in memory <b>106</b> and thus made available to processor <b>105</b>, the image is used to generate an index into the stored collection of documents (step <b>420</b>), from which an image of the corresponding stored reference document is retrieved and made available to processor <b>105</b> (step <b>430</b>).

0048Also, information about the active elements in the retrieved document image is obtained and is made available to processor <b>105</b> (step <b>440</b>). Such elements can be, for example, printed or typeset text or handwritten characters, pictures, graphics, photographs, or any other elements that make up the overall document. According to the invention, an active element need not be a blank element and need not contain any interior whitespace. This is in contrast with the active elements of typical forms of the prior art (such as check boxes <b>1</b> and rectangles <b>2</b> of the form of FIG. 1), which are typically blank lines, spaces, boxes, or other blank elements. In this embodiment, the active elements for each reference document are determined ahead of time and are stored in and retrieved from the collection of documents along with the reference document images. In some embodiments, the active elements can be separately determined after the reference document image has been retrieved.

0049Next, processor <b>105</b> extracts the user's mark from the marked document instance by performing reference-based mark extraction (step <b>450</b>). That is, processor <b>105</b> compares the obtained image of the marked document instance with the retrieved image of the reference document to determine what the user has changed in her or his version of the document. Preferably, robust techniques are used for this step, as will be discussed below, so that scaling errors, translation errors, transmission and digitization errors, noise, and other image artifacts can be tolerated.

0050Once the mark is extracted, processor <b>105</b> interprets the mark to determine what, if anything, the user wants done in response (step <b>460</b>). In this embodiment, the interpretation is based on the proximity of the mark with respect to active elements of the document. If the mark is not located near to any active element, the mark is ignored. If the mark is located near to an active element, the mark is interpreted in conjunction with that element. For example, if a straight-line mark or X mark or check mark appears underneath or through or over a word, or an oval mark appears around the word, the mark can be interpreted to mean that the user wants to do something pertaining to the word, such as selecting and following a hypertext link associated with the word or performing optical character recognition on the word and providing the result as input to a search engine, relational database, word processor, or other computational process. In other embodiments, the interpretation can be based on additional factors, such as the shape of the mark and a more sophisticated analysis of the geometric relationship between the mark and the nearby active element. Thus, for example, the interpretation of a line beneath a word can be distinguished from that of a line through a word. In still other embodiments, certain marks, such as a large X mark substantially the size of an entire page, can be interpreted independently of any particular active elements of the document.

0051Finally, processor <b>105</b> performs, initiates, or otherwise causes to be performed any action (or actions) indicated by the interpretation of the user's mark (step <b>470</b>). The action can be any arbitrary computation.

0052The action to be taken for each active element can be determined in advance and stored along with the active elements. For example, as previously stated, the processor can follow a hypertext link to retrieve a document related to an element of the document as indicated by the user's mark. The assignment of links to active elements can be done ahead of time. As another example, the processor can take or initiate an action represented by a graphical element in the document; e.g., if the user circles an icon representing a sprinkler, the computer can cause an automatic sprinkler system to be activated. However, the association of active elements with actions to be taken is entirely arbitrary. Thus, continuing the previous example, the action prompted by the user's selection of the sprinkler icon could be something entirely unrelated to sprinklers, such as causing an audio recording to be played or starting a word processing program.

0053Alternatively, for some or all elements, the action to be taken in response to a user's mark can be computed after the mark has been interpreted. This can be appropriate, for example, where the action to be taken depends on the document content, as in the example given earlier of performing optical character recognition on an underlined word (e.g., a name from a list of employee names) and then providing the recognized word to another program (e.g., a relational database program), which in turn can retrieve new documents (e.g., personnel records) and output them to the user.

0054If the indicated action calls for output of a document or documents, these can be output from computer <b>100</b> to any point in system <b>10</b>. For example, the document(s) can be output via printer <b>104,</b> or can be communicated from computer <b>100</b> to other devices in system <b>10</b> (such as fax machine <b>20</b> or multifunction device <b>30</b>) for output by those devices. As another example, the output document(s) can be electronically mailed from computer <b>100</b> to other computers across the Internet <b>50</b>.

0055It will be observed in the flowchart of FIG. 4 that image processing techniques are used for the mark extraction step, so that symbolic processing is not needed. Interpretation of the mark's significance takes place after extraction is complete. Although in some embodiments, the content of the document and the shape of the mark can be recognized (as for purposes of finding active elements in the document and/or assigning appropriate actions thereto), according to the invention they need not be recognized at all. For example, the computer can simply look for an element that the user has indicated is active, and (whatever that element is, and whatever the mark is) if the element is sufficiently near to the mark, the element is treated as having been marked and the computer takes or initiates the appropriate action.

0056FIG. 5 schematically illustrates the flow of information in the method of FIG. 4. A marked document instance <b>510</b>, obtained in step <b>410</b>, is used in step <b>420</b> to provide an index <b>511</b> into a database <b>520</b> or other collection of stored documents and so to retrieve a corresponding known reference document <b>521</b>. The pixel image <b>512</b> of marked document instance <b>510</b> is compared in step <b>450</b> with a pixel image <b>522</b> of reference document <b>521</b> to produce an extracted mark <b>550.</b> Also, activation information <b>524</b> is retrieved for reference document <b>521</b> (this information having been determined advance) and this supplies the active elements <b>540</b> of the document in step <b>440</b>. Given extracted mark <b>550</b> and active elements <b>540</b>, the user's mark can be interpreted in step <b>460</b> to produce interpretation <b>560</b>. This interpretation yields action <b>570</b> that is carried out in step <b>470</b>.

0057FIG. 6 is a subsidiary flowchart showing in greater detail the indexing step <b>420</b> of FIG. 4. Processor <b>105</b> generates an index (step <b>421</b>) and uses this index to find in the database or other collection of documents a stored image that corresponds to the index (step <b>423</b>). A pointer to the image or other designator as to where the image data is to be found is then returned (step <b>425</b>). so that the image can be retrieved (in step <b>430</b> of FIG. 4).

0058Preferably, and in this embodiment, an image-based indexing technique is used to generate the index in step <b>421</b>. An image-based index can be, for example, a number or set of numbers derivable from an image by performing certain image processing operations on that image. The number or numbers are defined such that visually similar documents will tend to have similar indices. Accordingly, an index generated from the image of the user's marked document instance will be close to an index generated from an image of the reference document from which the marked document instance derives.

0059In this embodiment, an index for each reference document image in the collection is precomputed and is stored in association with the image. Thus, in step <b>421</b>, the image processing operations needed to derive the index are performed on the image of the marked document instance. The documents of database 520 are stored together with their indices, which are precomputed, and a stored document having an index most closely matching the generated index is found in step <b>423</b>.

0060It is possible that two or more stored documents will have similar indices. In this case, a small set of documents, rather than a single document, is found in step <b>423</b>. Thereafter, as part of step <b>425</b> or step <b>430</b>, each image in the set can be retrieved and compared with the marked document instance image (e.g., by robust matching or by simple correlation) to determine the most appropriate reference document image from among the documents of the small set.

0061Various image-based indexing techniques can be used for Formless Forms. For example, wavelet decomposition techniques can be used. One such suitable indexing technique is disclosed in "Fast Multiresolution Image Querying" by Charles E. Jacobs et al. (University of Washington Technical Report UW-CSE-95-01-06), wherein it is stated (in section 1, italics in original) "[W]e define an <i>image querying metric</i> that makes use of truncated, quantized versions of the wavelet decompositions [of the images], which we call <i>signatures.</i> The signatures contain only the most significant information about each image." These signatures can be used as indices in the method of the present invention.

0062Alternatively, a brute-force image-based technique (not strictly speaking an indexing technique, but nonetheless workable) can be used to find the reference document. In this technique, each document in the collection is compared with the marked document instance, using a robust image comparison technique such as Hausdorff matching. Exhaustive search by this approach will eventually lead to the reference document, although this is less efficient than indexing.

0063In still another approach, a symbolic indexing technique can be used to generate the index in step <b>421</b>. This approach depends on there being symbolic content in the document. For example, if the document contains printed text, optical character recognition can be performed on the textual portions of the document and the resulting characters used to generate a set of hash codes. The set of hash codes, in turn, is used as the index of the document. As another example, if the document happens to be already coded with a machine-readable symbolic code, such as a data glyph or bar code, this machine-readable code can be used as the index of the document.

0064Persons of skill in the art will appreciate that a number of different techniques can be used to retrieve the appropriate reference document from a collection of such documents given a marked document instance. Image-based techniques have great generality, and do not require the indexed documents to include any particular kind of content. Symbolic techniques can be used where and as appropriate.

0065FIG. 7 is a subsidiary flowchart showing in greater detail the reference-based mark extraction step <b>450</b> of FIG. 4. Extraction proceeds in two basic steps: registration and differencing. In registration (step <b>451</b>), processor <b>105</b> determines how pixels in the marked document instance image map onto pixels in the reference image. In differencing (step <b>452</b>), the pixels that represent the user's mark are isolated within the image. Essentially, processor <b>105</b> looks for every element in the image of the marked document instance that matches a corresponding element in the reference document image, and eliminates all these matches; whatever is left over is presumed to be something that the user added, that is, a mark.

0066Registration is performed because the marked document instance image can be translated, rotated, scaled, skewed, distorted, and otherwise misaligned with respect to the reference document image. If differencing is performed on unregistered images, it is likely to produce spurious results. Preferably, registration is performed using Hausdorff matching techniques, although straightforward binary correlation techniques could also be used in principle (Notably, registration does not rely on the use of registration marks or other special symbols, and thus can be accomplished in the image domain, without symbolic processing.)

0067In differencing, processor <b>105</b> determines the elements that are common to both images (step <b>453</b>), finds the user's mark or marks by eliminating the common elements (step <b>455</b>), and then optionally performs thresholding or other "clean up" operations to eliminate noise, unintentional stray marks, and the like, thus leaving the meaningful mark or marks (step <b>457</b>). Preferably, differencing is performed on the registered images using robust image differencing techniques, although straightforward binary correlation techniques could also be used in principle.

0068In this embodiment, a robust image differencing technique is used. In step <b>453</b>, the common elements are found by applying a matching technique to the registered images. Any group of black pixels (assuming here that the images are black-on/white-off binary images) that is found in both of the registered images and whose respective positions in the two registered images are sufficiently "close" to one another are deemed to be common elements (that is, elements of the marked document instance image which are adequately accounted for by elements of the reference document image and which, therefore, presumably were not introduced by the user's mark.) In step <b>455,</b> the complement of the matching result obtained in step <b>453</b> is taken, thus eliminating the common elements and leaving only those elements that were unmatched. Put another way, in step <b>455</b> the processor <b>105</b> finds groups of black pixels in the marked document instance image that have no counterparts in the reference document image. Thereafter, in step <b>457,</b> a thresholding of connected components according to component size can be performed to eliminate isolated noise pixels and the like. This leaves only larger, substantially contiguous groups of black pixels to be considered as marks.

0069The robust differencing performed in steps <b>453</b> and <b>455</b> can be expressed more formally as follows: Let <i>M</i> = {<i>m</i>} = {<i>m</i>_<i>1</i>, <i>m</i>_<i>2</i>, <i>m</i>_<i>3</i>...} be the set of black pixels in the registered reference image, and let /={<i>i</i>}= {<i>i</i>_<i>1</i>, <i>i</i>_<i>2</i>, <i>i</i>_<i>3</i> ...} be the set of black pixels in the registered marked instance image. Pixels m in M have x and y coordinates (x_<i>m</i><i>,</i> y_<i>m</i>) and pixels <i>i</i> in I have x and y coordinates (x,. y,); for example, pixel <i>m</i>_<i>1</i> has coordinates (x_<i>m</i>_{<sub2><i>1</i></sub2>}, y_<i>m</i>_{<sub2><i>1</i></sub2>}) and pixel <i>i</i>_<i>1</i> has coordinates (x_{<i>i</i><sub2><i>1</i></sub2>}, y_{<i>i</i><sub2><i>1</i></sub2>}). Compute<maths id="math0001" num=""><img file="EP0805410B1_D0001.tif" /></maths> for all i in <i>I;</i> that is, Δ(<i>i</i>) = the minimum distance between pixel i and any pixel in <i>M.</i> Then the set of mark pixels <i>Y</i> is<maths id="math0002" num=""><math display="block"><mrow><mtext mathvariant="italic">Y</mtext><mtext> = {</mtext><mtext mathvariant="italic">i</mtext><mtext>|Δ(</mtext><mtext mathvariant="italic">i</mtext><mtext>) > δ}</mtext></mrow></math><img file="EP0805410B1_D0002.tif" /></maths> where δ is a selected threshold value. The set <i>Y</i> optionally can be further processed (as in step <b>457</b>) to eliminate mark pixels that are likely to be due to noise or otherwise insignificant.

0070To accomodate nonlinear distortions, piecewise computation of the registration and differencing steps can be used. This approach will be helpful if pieces are sized such that that locally within any piece, the distortion is not too drastic. The piecewise computation is illustrated schematically in FIGS. 8A-8B. In each of FIGS. 8A and 8B, an image <b>80</b> is divided into local regions (regions <b>81</b> in FIG. 8A and regions <b>81'</b> in FIG. 8B), and each region is separately registered and differenced. The regions can be nonoverlapping, as shown in FIG. 8A, or can overlap, as shown in FIG. 8B.

0071When using the piecewise approach, the pieces are independent of one another, so that the loop order implied by FIG. 7 can be reversed: That is, it is possible either to compute registrations for all pieces, then compute differencings for all pieces; or to compute registration and differencing for the first piece, registration and differencing for the second piece, etc. It will be appreciated that other approaches, such as processing the various pieces in parallel (as where the "processor" is in fact a multiprocessor architecture or a collection of computers operating in parallel), are also possible.

0072Note that in the differencing step <b>452,</b> the <i>entire</i> marked document instance image is compared with the <i>entire</i> reference image, whether or not differencing is accomplished piecewise as shown in FIGS. 8A-8B. This contrasts with known computerized form interpretation methods in which only predesignated active regions of a form are analyzed and other areas are ignored for efficiency. For example, in a conventional method, if a form includes a paragraph of explanatory text and a set of check boxes, the check boxes are the only elements of concern to the computer. So the computer focuses its analysis of the marked-up form in those areas of the page where the check boxes appear and ignores the rest of the page, since any analysis of portions of the form beyond those necessary to determine whether the check boxes have been marked would be wasteful of computing resources. By contrast, according to the invention, comparison of the entire marked-up form with the entire reference document is advantageous, not wasteful. It is, in part, why arbitrary documents can be used as Formless Forms: A mark can be made and recognized <i>anywhere</i> in a document, because the computer looks for marks <i>everywhere</i> in the document.

0073FIG. 9 is a flowchart showing how the active elements of a document image (such as the retrieved document image obtained by processor <b>105</b> in step <b>440</b>) can be designated. In this embodiment, the active elements of reference document images are designated in advance and stored with the document images in database <b>520</b>, so that the active elements can be retrieved contemporaneously with the reference document image. In other embodiments, the active elements can be stored separately from the reference document images, or can be designated or otherwise determined at run time.

0074To designate the active elements, the reference document image is scanned, retrieved from storage, or otherwise made available to processor <b>105</b> (step <b>901</b>). Thereafter, active elements are designated (step <b>903</b>) and are linked (that is, associated) with their respective actions (step <b>905</b>). The designation and linking can be done by a user with appropriate editor software, wherein the user manually designates which elements of the image are to be active and what actions are to be associated with each. Alternatively, designation and linking can be done semiautomatically, as where the processor carries out symbolic recognition of the image content (e.g., OCR plus word recognition), designates each recognized symbol as an active element, and assigns an action according to the type of symbol and the recognized content. The active elements and their associated actions are stored for later use (step <b>907</b>) as a lookup table, association list, or other suitable data structure.

0075In FIG. 9, the reference document is referred to as a "legacy" document, that is, a document that predates the Formless Forms system (here, system <b>10</b>) in which it is used. Typically, a legacy document is a document that was not originally intended for use as a form For example, it can be an original or photocopied paper document that has been scanned into system <b>10</b>. According to the invention, a legacy document can nevertheless be used as a reference document of the collection. Thus a legacy document that was not designed for use as a form can nevertheless be used as a Formless Form. Moreover, a legacy document can be converted into a Formless Form without symbolic recognition of the legacy document image, and the designation and assignment of active elements in the document and actions to be taken upon a user's selection of such elements can be essentially arbitrary.

0076The series of views of FIGS. 10-12 show exemplary test results produced by reference-based mark extraction software suitable to carry out the steps of the flowchart of FIG. 7, using piecewise registration and differencing as was illustrated in FIG. 8. The software is written in Lisp and in C++ and was run on a Silicon Graphics Indy workstation. FIG 10 shows a portion of the reference document image. FIG. 11 shows the marked document instance, which in this example is a marked-up fax copy of the reference image. FIG. 12 shows the extracted marks.

0077The series of views of FIGS. 13-19 show an illustrative example of processing a Formless Form according to the steps of the method of FIG. 4. FIG. 13 shows the original document <b>700,</b> which will be used as the reference document in this example. Document <b>700</b> includes various elements surrounded by a whitespace background <b>701.</b> The elements of document <b>700</b> include handwritten words such as words <b>710, 711,</b> and <b>712,</b> typeset text <b>720</b>, hand-drawn graphical elements including flower <b>731</b>, face <b>732</b>, and spade <b>733,</b> typeset graphic element <b>740,</b> and a detailed graphic <b>750</b> from a U.S. Postal Service first-class stamp. (Note that a given group of contiguous or related black pixels can be considered a single element for some purposes and a collection of elements for other; thus, for example, typeset text element 720 can also be considered to be a collection of many separate elements such as letter characters <b>720a</b>, <b>720b</b> and number characters <b>720c</b>, <b>720d</b>.)

0078FIG. 14 shows an instance <b>800</b> of document <b>700</b>. The user has instance <b>800</b> on hand and wants to use as input to a Formless Forms PUI that has document <b>700</b> among its collection of reference documents. Instance <b>800</b> is a photocopy of document <b>700</b> and is reduced in scale from the original. It is unmarked.

0079FIG. 15 shows an instance <b>800'</b> of document <b>700</b>. It is the same as document instance <b>800</b> of FIG. 14, except that it has been marked by the user with marks <b>810</b> and <b>820</b>.

0080FIG. 16 shows an instance <b>800"</b> of document <b>700.</b> It is the marked document instance as received via fax (e.g., received by computer <b>100</b> from fax <b>20</b>) and provided to processor <b>105</b> (step <b>410</b>). Note the stairstepping artifacts ("jaggies") caused by the relatively low resolution of the fax. These are visible, for example, in the bottom portion of mark <b>810'</b> (the faxed version of user mark <b>810</b>).

0081Processor <b>105</b> generates an image-based index from the image of instance <b>800"</b> into the stored collection of documents (step <b>420</b>), from which an image of the corresponding stored reference document, that is, document <b>700,</b> is retrieved (step <b>430</b>). Also, information about the active elements in the retrieved document image is obtained by processor <b>105</b> (step <b>440</b>). FIG. 17 shows an exemplary designation of active elements for document <b>700.</b> Dashed rectangles, such as rectangles <b>711a</b> and <b>731a</b>, have been drawn around elements <b>710, 711, 712, 731,</b> and <b>750</b> to indicate schematically that these elements have been designated as active.

0082The dashed rectangles in FIG. 17 delimit the extent of the elements they surround. This can be done in two ways: (1) An entire rectangle can be defined as an active region. A user mark near, in, or substantially overlapping this region is deemed to pertain to the rectangle, and thus to the active element surrounded by the rectangle. Note that with this approach, a rectangular region can be drawn anywhere in the image, and need not surround an element; for example, a rectangle could be drawn so as to encompass only the right-hand half of stamp <b>750</b> or only the upper half of flower <b>731</b>. (2) The rectangle indicates a group of black pixels that are to be treated as a single element, regardless of whether these pixels are contiguous. The actual boundaries of the element, however, are not necessarily rectangular, but are determined by the shape of the element itself, e.g., by dilation of the boundary of the element. This approach is helpful in cases where it is not immediately clear whether a group of black pixels is to be treated as a single element or as multiple elements (as was discussed previously with respect to element <b>720</b> in FIG. 13, for example).

0083Thus the dashed rectangles <b>711a</b> and <b>731</b> can be seen either as themselves being the boundaries of their respective active elements <b>711</b> and <b>731</b> (approach (1)), or as simply being indicative of the general vicinity in which the actual boundaries, as determined by the element shape, are located (approach (2)). The latter approach (approach (2)) is preferably used in this embodiment. Whatever way they are defined, the location of active element boundaries can be important in determining which mark pertains to which active element in the mark interpretation step (step <b>460</b>).

0084In general, in order to establish the boundaries of particular active elements, an active link editor program (step <b>903</b> of FIG. 9) can be run on the image of document <b>700</b>. Methods are known for characterizing active elements in legacy documents, either manually or semiautomatically (e.g., methods for converting legacy documents into Web or other hypertext documents). These methods and similar methods can be applied in such an editor program. For example, an editor program can be made to run interactively on a computer having a graphical user interface. The editor program presents an image of document <b>700</b> and allows the person running the program to designate elements to be activated by selecting these elements with an editing tool that establishes rectangular or other desired boundaries around particular pixel groupings. In particular, rectangular boundaries corresponding to the dashed rectangles of FIG. 17 can be selected for the elements of document <b>700</b>.

0085FIG. 18 shows a difference image <b>880</b> containing extracted marks <b>810", 820".</b> It is the result of performing mark extraction (step <b>450</b>) on image <b>800"</b> using the image of document <b>700</b> as the reference image. (The image of FIG. 18 is, with respect to the example of FIGS. 13-19, what the image of FIG. 12 is with respect to the example of FIGS. 10-12.)

0086FIG. 19 shows an image <b>890</b> that includes the extracted marks <b>810", 820"</b> from difference image <b>880</b> plus the elements <b>731", 711"</b> to which these marks respectively pertain Although image <b>890</b> need not explicitly be formed during the interpretation step (step <b>460</b>), it is illustrative of an interpretation result, and is intended here to schematically depict such a result. The interpretation is that the user has selected two of the five active elements of document <b>700</b>, namely, word <b>711</b> and flower <b>731</b>. Processor <b>105</b> can then take any actions associated with these selected elements. (step <b>470</b>).

<i>Paper Web</i>

0087The World Wide Web has become the most active part of the Internet. The Web has a large and rapidly growing number of sites, each site typically offering a collection of hypertext documents, or pages. (For purposes of this description, it is convenient to speak of each Web page as being a single hypertext document, and to treat a collection of interlinked Web pages as being a database of hypertext documents, rather than as being a single larger hypertext document.) Each page in the Web has a unique identifier called a universal resource locator, or URL. The Web as a whole can be thought of as a heterarchically organized database of hypertext documents.

0088Web pages are active documents. They can be linked to each other through hypertext links (also known as "hot links," active links, hyperlinks, etc.). A link provides a reference from one Web page to another. Each link associates a displayed or displayable element of a Web page in which the link appears with a URL that identifies another Web page. When the user selects the displayed element, the computer retrieves the associated URL and then uses the retrieved URL to retrieve the Web page that the URL identifies. Thus the user can move through the Web from one page to another to another by following the links on successive pages.

0089Web pages supporting textual, graphical, photographic, audio, and video materials are written in hypertext markup language (HTML). Web pages can also be written to support executable applications software, such as software written in the Java programming language (available from Sun Microsystems, Inc., Mountain View, CA).

0090The Web is well suited for interactive, on-line use. Typically, a Web user browses (or "surfs") the Web by running browser (client) software on a computer equipped with a graphical user interface. The brower software can be an applications program, such as Netscape Navigator (available from Netscape Communications, Inc., Mountain View, CA), or can be built into the operating system. The browser lets the user display Web pages on a visual display device, such as a CRT or flat-panel display screen. The browser also lets the user select and follow links in the Web page with a pointing device, such as a mouse or a trackpad. Selecting a displayed link causes the computer to traverse the link, that is, to fetch the Web page pointed to by the URL associated with the link.

0091Known Web browsers can render an HTML document as a pixel image and can instruct the computer about what portions of rendered image are "hot links,' that is, are to be treated as active elements that can be addressed by the user with a mouse or other pointing device so as to prompt the computer to retrieve other Web pages. Known Web browsers can also print out the rendered images of Web pages. Unfortunately, when the page is printed, the hypertext link information is ordinarily lost.

0092The question, then, is how to restore the lost link information to the printed Web page. This can be done by treating a printed Web page as a Formless Form. The result is Paper Web.

0093Paper Web is conceived as a new kind of Web browser It, too, can render an HTML document and determine the positions of active elements in the rendered image Paper Web creates and saves a map of the active elements, their positions in the rendered image, and their associated hypertext links. PaperWeb can then treat a printout of the rendered Web page as a Formless Form, by using its stored map to restore the hypertext links that would otherwise be lost.

0094Here is an example of how Paper Web can work: Suppose the user is situated at a fax machine located remotely to a host computer that is in connection with the Web. The computer runs software to support a Paper Web browser having an integrated fax server. The computer retrieves a Web page and faxes a hardcopy of the retrieved page to the user. The user marks the hardcopy to indicate a particular hypertext link that the user wishes to follow. For example, the user circles, underlines, or draws an X over a graphical object, text string, or other active element representing the link The user then faxes the hardcopy thus marked back to the computer. The computer determines what Web page the user has sent and what Web link the user has indicated. The computer then follows the indicated link to obtain a new Web page, which it faxes back to the user.

0095FIG. 20 shows an example of a printed Web page <b>1000</b> which has been marked by the user and faxed back to computer <b>100</b> for processing by Paper Web. The user has marked a graphical button <b>1001</b> with an X mark <b>1002,</b> indicating that the hypertext link represented by button <b>1001</b> is to be followed.

0096The series of views in FIG. 21 illustrates an example of a sequence of interactions between a user and Paper Web. In view (1), the user inputs an instance of (e.g., scans or faxes into the computer a paper copy or printout of) a page <b>1100</b> on which the user has indicated two links by writing X marks on active elements <b>1101, 1102</b> of page <b>1100</b>. Paper Web interprets these marks and follows the links to retrieve two new pages <b>1110, 1120</b>, which are printed out for the user. Paper Web also adds the new pages <b>1110, 1120</b> to its collection of known documents. (Note that with Paper Web, the user can follow multiple links at the same time, something not ordinarily possible with typical known GUI-based Web browsers.)

0097In view (2), the user inputs a marked instance <b>1110'</b> of the page <b>1110</b> that was previously retrieved in view (1). The user has indicated by X marks on active elements <b>1111, 1112</b> two links to be followed from page <b>1110</b>. Paper Web recognizes page <b>1110</b> and calls up its stored image from the collection of reference documents. Then, using the stored image of page <b>1110</b> to extract and interpret the user's marks, Paper Web follows the indicated links to retrieve two additional new pages <b>1130, 1140,</b> which are printed out for the user and which are also added to the collection.

0098In view (3), the user inputs a marked instance <b>1120'</b> of the page <b>1120</b> that was previously retrieved in view (1). The user has indicated by an X mark on active element <b>1121</b> a link to be followed from page <b>1120</b>. Paper Web recognizes page <b>1120</b> and calls up its stored image from the collection of reference documents. Then, using the stored image of page <b>1120</b> to extract and interpret the user's marks, Paper Web follows the indicated link to retrieve another additional new page <b>1150,</b> which is printed out for the user and which is also added to the collection.

0099Finally, in view (4), the user inputs an instance of the user's own home page <b>1090</b> Paper Web is assumed to keep this page (or, more generally, at least one default page representing a convenient starting point for the user) in its collection at all times. The user has indicated by an X mark on active element <b>1091</b> a link to be followed from page <b>1110</b>. Paper Web recognizes page 1090 and calls up its stored image from the collection of reference documents. Then, using the stored image of page <b>1090</b> to extract and interpret the user's marks, Paper Web follows the indicated link to retrieve yet another additional new page <b>1190</b>, which is printed out for the user and which is also added to the collection.

0100FIG. 22 is a high-level flowchart for the method of Paper Web in one embodiment. These steps are carried out using appropriate components of system <b>10</b> and, more particularly, computer <b>100</b> under control of processor <b>105.</b>

0101Initially, the user invokes Paper Web, for example, by calling computer 100 on the telephone or by scanning in a special start-up document. This causes Paper Web to output a hardcopy of the user's default page or pages (e.g., the user's home page or the user's "bookmark list" containing links to the user's favorite Web sites) to the user (step <b>2200</b>).

0102Thereafter, Paper Web enters a work loop in which the user marks up instances of hardcopy Web page printouts in his or her possession (step <b>2210</b>) and sends these instances to the computer, which follows the indicated links and sends back new printouts. The computer continually updates a cache of the Web pages previously visited by the user, so that these pages can be recognized if marked instances of the pages are presented as input.

0103More particularly, the loop begins with the user marking one or more links on a paper or other hardcopy instance of a Web page known to Paper Web (step <b>2210</b>). A scanned, faxed, or other pixel image of this marked page instance is provided to system <b>10</b>, sent to computer <b>100</b>, and stored in memory <b>106</b> for use by processor <b>105</b> (step <b>2220</b>).

0104Once the image of the marked page instance has been made available to processor <b>105</b>, the image is used to find the corresponding known Web page in the cache (step <b>2230</b>). This can be done, for example, by generating an image-based or other index from the marked page instance image. The cache stores associations between the indices of previously visited Web pages (i.e., pages known to the system and therefore usable as reference documents) and the contents of these pages. The cache can store the actual contents of each cached page (that is, the HTML representation of the page), or simply store the page's URL.

0105After the correct page has been found, processor <b>105</b> retrieves the image and hypertext link information for the page (step <b>2240</b>). If the cache contains the HTML representation of the page, the processor can render the HTML directly to produce a bitmap and to determine where the active regions are in this bitmap. If the cache contains only the URL, the processor first can fetch the page from a Web server storing the reference documents.

0106Next, processor <b>105</b> extracts the user's mark from the marked document instance by performing reference-based mark extraction (step <b>2250</b>), using as the reference document image the rendered image of the cached page. That is, processor <b>105</b> compares the image of the user's marked page instance with the rendered image of the cached page to determine what links the user has marked. Preferably, Hausdorff registration and robust differencing techniques are used to accomplish this step, as described earlier in connection with Formless Forms.

0107Once the mark is extracted, processor <b>105</b> interprets the mark or marks to determine what links, if any, the user has selected (step <b>2260</b>). This can be done, for example, by determining the proximity of the user's mark or marks to the pixels that represent the hypertext links of the page. In other words, Paper Web interprets the user's mark or marks by treating each mark as the paper-based equivalent of a mouse click. Just as a GUI-based Web browser can determine the location of a user's mouse cursor with respect to a rendered image of a Web page and thereby can determine what link, if any, the user has selected, here Paper Web can determine the location of the user's mark or marks with respect to the rendered image of the Web page and thereby can determine what link or links (if any) the user has selected. Thus, if the pixels of the user's mark intersect (or are nearby to, or otherwise indicate) pixels of the rendered Web page that correspond to Web page elements representing one or more hypertext links, these links are deemed to have been selected by the user.

0108Processor <b>105</b> follows any link or links that the user has the selected and retrieves the Web pages indicated by these links (step <b>2270).</b> Processor <b>105</b> computes index values for these pages and saves the index values, together with the page's HTML contents or URL as the case may be, in the cache (step <b>2271</b>). In this way, the newly retrieved pages will be recognizable if, at a future time, the user wishes presents them to Paper Web in hopes of revisiting them. Also, processor <b>105</b> provides hardcopy output of the retrieved pages to the user (step <b>2272</b>). At this point the loop can either continue (step <b>2280</b>) or terminate, in which case Paper Web is exited (step <b>2290</b>).

0109Paper Web maintains a continually growing cache of known pages in this embodiment. In other embodiments, the cache can be limited in size and be purged from time to time of little-used pages. At a minimum, at least one page, such as the user's home page or other default page, should remain in the cache at all times, so that Paper Web always has at least one page that it can recognize and that the user can use as a starting point for Web exploration.

0110FIG. 23 is a subsidiary flowchart showing in more detail how, in step <b>2240</b> of FIG. 22, Paper Web determines the locations of hypertext links in retrieved pages. Paper Web begins by rendering the HTML representation of the page to produce a rendered pixel image (step <b>2241</b>). Next. Paper Web determines which pixels of the rendered image correspond to the active elements specified in the HTML (step <b>2243</b>). This can be done, for example, using methods similar to those known for solving the analogous problem in GUI-based Web browsers. That is, Paper Web can determine the locations of those pixels corresponding to hypertext links for a rendered image of a given Web page in much the same way that a GUI-based Web browser of the prior art can determine the corresponding pixel locations for a displayed image of that Web page. Once the pixels representing active elements in the rendered image have been located, the links can be obtained by associating these active elements with their corresponding URLs, as specified in the HTML representation of the page (step <b>2245).</b>

0111Finally, the map of active links is stored (step <b>2247</b>) for later use during the mark interpretation step (that is, in step <b>2260</b> of FIG. 22). However, the map need not be stored with the cached page. Whereas in the embodiment of Formless Forms described earlier, the map of active links for each reference document was preferably precomputed (as shown in FIG. 9) and stored with the reference document, here, the map of active links can conveniently be computed at run time. That is because unlike legacy documents, which must be specially converted for use as active or hypertext documents, HTML documents are intended from the outset to be hypertext documents. So, once Paper Web has found the appropriate cached page corresponding to a given input page instance, Paper Web can readily regenerate the map of pixels to links from the HTML representation of the page. Paper Web, in effect, restores the hypertext functionality that otherwise is lost upon printing out the Web page.

<i>Conclusion</i>

0112Formless Forms provides a way to "bring paper documents to life" by turning ordinary, inactive/nonhypertext documents into active forms and hypertext documents. The Formless Forms approach can be used, for example, to turn legacy documents that were never intended for use as forms into forms, or (as in Paper Web) to make paper printouts of World Wide Web or other hypertext pages as powerful and useful as their on-screen counterparts. With Formless Forms and Paper Web, computer users can access the World Wide Web or any hypertext database without a mouse, keyboard, or display screen. All the user needs is a fax machine, a pen, and a telephone number to call their computer.

0113Four key features of Formless Forms are: <ul id="ul0002" list-style="none" compact="compact"><li>1) <i>indexing</i>    (e.g., image-based indexing of a marked document instance to retrieve a corresponding reference document from a database);</li><li>2) <i>reference-based mark extraction</i>    (e.g., registration of marked and unmarked document images via Hausdorff or other matching techniques, followed by robust differencing of the registered images);</li><li>3) <i>designation</i> and 4) <i>interpretation</i> of active elements in what would otherwise be inactive documents    (e.g., associating textual, graphical, photographic, or any other elements of legacy documents with corresponding actions, thus allowing even non-forms to behave like forms; associating elements of hardcopy versions of Web or other hypertext documents with corresponding hypertext links in the original HTML or other source documents, thus allowing even a paper printouts of a Web page to provide the user with ready access to any and all other linked Web pages).</li></ul>

0114In Formless Forms, the appearance of the page defines the form, rather than vice versa. Thus the appearance of a computer-recognizeable form for a PUI need no longer be restricted as in the past. A "form" can be made from virtually any document―whether or not that document was ever intended for use as a form―and its design need not be confined by the limited graphical vocabulary of known form editors and design tools. Moreover, the time and effort invested in designing new machine-readable forms can be reduced, because legacy documents (in particular, printed forms designed to be read by human beings but not by machines) can be made machine-readable.

0115The foregoing specific embodiments represent just some of the possibilities for practicing the present invention. Many other embodiments are possible within the spirit of the invention. For example: <ul id="ul0003" list-style="none" compact="compact"><li>· The method of Paper Web can also be used with other hypertext document databases besides the World Wide Web. Such hypertext document databases include, for example, internal use of the Web within companies (so-called "Intranet") and CD-ROM based hypertext databases such as CD-ROM encyclopedias, collections of patents on CD-ROM, and the like. (In system <b>10</b>, CD-ROM <b>109</b> can be used to this end.)</li><li>· There are other kinds of Web page interaction devices in use besides simple active links. For example, many Web pages have type-in boxes (e.g., as for entering a user name or a search string), and some have point-and-click image panels (e.g., as for a graphical map browser, in which the user can point to a map location and the Web server responds by returning a more detailed map of the indicated location). For type-in boxes, the mark interpretation component of Paper Web can be augmented with handwriting recognition software to convert a user's hand-printed response (which is initially found, without recognition, by Paper Web's mark extraction component) to an ASCII text string that can be sent by the computer to the Web server. For a point-and-click image panel, the mark interpretation component of Paper Web can be augmented to determine relatively precise geometric information from the marked-up image. For example, if the user draws an X over the position of interest, the mark interpretation processing can find the intersection of the two lines that make up the X, and send the position of that intersection on to the Web server.</li></ul>

0116Accordingly, the scope of the invention is not limited to the foregoing specification, but instead is given by the appended claims.