Method and system for estimating the robustness of algorithms for generating characterizing information descriptive of a selected text block

0001The subject invention relates to the problem of providing a robust, compact characterization of a block of printed text which will distinguish the selected block of text from other such blocks. More particularly, it relates to the problem of estimating the robustness of algorithms for generating characterizing information descriptive of printed material. (By "robust and compact" herein is meant information which is small enough in quantity to be incorporated into postal indicia yet will identify a text block, and distinguish it from other text blocks, with sufficient reliability to deter "rubber stamp" counterfeiting; despite errors introduced by the printing and/or scanning processes.)

0002Postage metering systems account for postage and other values such as parcel delivery service charges and tax stamps, and print indicia representative of such values as proof of payment. To protect against counterfeiting of indicia, modern digital postage metering systems use encryption technology. The postage value and other information relating to an indicium are preferably digitally signed, or otherwise cryptographically authenticated, and the information and signature are incorporated into the digital postal indicium.

0003Digital postal indicia using encryption technologies are extremely secure. In general, without knowledge of the proper encryption keys, it is essentially impossible to produce a counterfeit digital indicium, However, digital indicia are subject, as are all postal indicia, to "rubber-stamp" counterfeiting when a valid indicium is scanned and reproduced on multiple mail pieces. To prevent such "rubber-stamp" counterfeiting, it is known to incorporate information from the address block of the mail piece into the postal indicium. Because space on an envelope is limited, typically only a small portion of the information in the address block will be incorporated into the indicium.

0004<b>In Figure 1, prior art mailing system 10 includes address printer controller</b> 12, address printer 14, postage meter 16, and indicia printer 20, Address printer controller 12 receives address information from a data processing system (not shown), generates a bitmap representative of the nominal, or "pristine", image of the address block, and controls address printer 12 to print address block A, representative of the address, on envelope E. Meter 16 receives postage information, and other information, from the data processing system. Meter 16 also receives characterizing information descriptive of block A from address printer controller 12. The information received can be either text-based or image-based. Text-based information is descriptive of the words or characters making up to the address, (e,g., ASCII code) while image-based information is descriptive of the actual printed image in the address block. Meter 16 combines the characterizing information with the postage value and other information, typically digitally signs the combination, generates a bitmap representative of an indicium including the digitally signed combination, and controls indicia printer 20 to print indicium IN on envelope E. When the mail piece is sent to a postal service location, the address block can be scanned again, and the information regenerated from the scanned address block compared to information recovered from indicium IN, without the need to communicate with the remote mailing system; thus tying indicium IN to the particular mail piece. (Note that since indicium IN is cryptographically linked to the address on the mail piece, printer 20 need not be a secure printer; but can be a general purpose printer which can be controlled by other devices for other uses.) Commonly assigned, <patcit id="pcit0001" dnum="US20040128254A" dnum-type="L"><text>U.S. patent application publication number 2004/0128254</text></patcit>, entitled <i>System And Method For Mail Destination Address Information Encoding Protection And Recovery In Postal Payment</i> in the name of Leon A. Pintsov /discloses a system similar to that of the Figure 1 using text-based characterizations of the address block.

0005While useful for its intended purpose, problems remain with the system of Figure 1 and similar systems. It has proven difficult to reliably recover textual information from address blocks during the validation process using available optical character recognition (OCR) techniques. Attempts to increase the robustness of text-based systems by incorporation of additional information and/or the use of error correcting codes has resulted in undesirable increases in indicia size and computational complexity. Use of image-based characterizing systems, such as those described in the above mentioned publication number 2004/0128254, has been proposed and is believed to substantially overcome some of the problems of text-based systems; however, it has proven difficult to form a priori estimates of the robustness of proposed characterizing algorithms for image-based system, forcing users to undertake extensive trial and error testing of various algorithms. This problem is exacerbated by the variation in robustness of characterizing algorithms with respect to the particular text block to be characterized.

0006<patcit id="pcit0002" dnum="US20010040979A1"><text>US 2001/0040979 A1</text></patcit> discloses that watermark and pattern detection can be improved by compensating for artifacts introduced into an image by a printer and/or scanner through which the image has passed. Prior to watermark or pattern detection, the image is filtered or modified to compensate for artifacts introduced by the printer and/or scanner. Some scanners automatically compensate for artifacts introduced by the scanner by using a calibrated tone map. The automatic compensation provides an image from which, a watermark can be easily read. However, generally the user is provided with an interface which can be used to change certain parameters such as contrast and intensity. The changes made by the user change the compensation (i.e. the tone map) applied to the image. If the user changes the compensation applied to the image it can affect the ability to read the watermark. A system is disclosed which reverses any compensation introduced by the user so that the watermark or pattern can be more easily read. Another system takes into consideration that some printers and scanners have transfer functions which differ in the "x" and "y" directions. Thus, the compensation introduced by the filter can differ in the "x" and "y" directions. In one system, a scanner introduces aliasing frequencies into an image. Detection is improved by selectively removing certain frequencies. In another system, the filter compensates for fact that the scanner frequency response falls off at higher frequencies.

0007<patcit id="pcit0003" dnum="US20030053653A1"><text>US 2003/0053653 A1</text></patcit> discloses a watermark system including an embedder, detector, and reader. The watermark embedder encodes a watermark signal in a host signal to create a combined signal. The detector looks for the watermark signal in a potentially corrupted version of the combined signal, and computes its orientation. Finally, a reader extracts a message in the watermark signal from the combined signal using the orientation to approximate the original state of the combined signal. While adapted for images, video and audio, the watermark system applies to other electronic and physical media. For example, it can be applied to mark graphical models, blank paper, film and other substrates, texturing objects for ID purposes, etc.

0008In accordance with the subject invention, there is provided a method for selecting a characterizing algorithm for generating a characterizing information descriptor for a selected block of printed material where the printed material is to be scanned from an object and compared with the characterizing information descriptor at a location distant from where the block is printed. The method comprises: printing the block on an object; applying each algorithm from a predetermined set of characterizing algorithms to a pristine image of the block of printed material to generate a plurality of corresponding first characterizing information descriptors for the block; determining estimates of robustness, with respect to the block of printed material, for each of the algorithms in the set to determine which of the characterizing algorithms is most robust; and selecting a descriptor generated by the algorithm and being so determined to be most robust to be used at the distant location.

0009In an embodiment, the estimates are determined by filtering the pristine digital image of the block of printed material with a print/scan filter to create a filtered image, the print/scan filter simulating the expected transformation of the pristine image by printing and scanning processes; applying each algorithm from the predetermined set of characterizing algorithms to the filtered image to generate a plurality of corresponding second characterizing information descriptors for the filtered digital image; and, for each algorithm from the predetermined set of characterizing algorithms, comparing corresponding the first and the second descriptors to determine which of the characterizing algorithms is most robust.

0010In an embodiment, the object is a mail piece and the block of printed material represents an address and the selected descriptor is comprised in an indicium printed on the mail piece; whereby the selected descriptor can be recovered from the indicium for use at the remote location.

0011In an embodiment, the selected descriptor is one of the second descriptors.

0012In an embodiment, the estimates are determined by filtering the pristine digital image of the block of printed material with a print/scan filter to create a filtered image, the print/scan filter simulating the expected transformation of the pristine image by printing and scanning processes; further filtering the filtered image with one or more defacing filters, the defacing filters simulating simulate blots, smudges, failure of print elements or scanner sensors, or other, similar occasional events which can not easily be incorporated into the print/scan filter to create one or more defaced images; applying each algorithm from the predetermined set of characterizing algorithms to the filtered image and to the one or more defaced images to generate a plurality of corresponding second characterizing information descriptors for the filtered digital image and one or more pluralities of defaced image descriptors corresponding to each of the one or more defaced images; and for each algorithm from the predetermined set of characterizing algorithms, comparing corresponding first characterizing information descriptors with corresponding second characterizing information descriptors and with each of the one or more corresponding defaced image descriptors to determine which of the characterizing algorithms is most robust.

0013An embodiment of the present invention provides a method and system for estimating the robustness of characterizing algorithms with respect to a particular text block.

0014Objects and advantages of embodiments will be apparent to those skilled in the art from consideration of the detailed description set forth below and the attached drawings.

0015The present invention is illustrated by way of example, and not by way of limitation, in the figures of the accompanying drawings and in which like reference numerals refer to similar elements and in which: <ul id="ul0001" list-style="none"><li>Figure 1 shows a schematic block diagram of a prior art mailing system.</li><li>Figure 2 shows a schematic block diagram of a mailing system in accordance with an embodiment.</li><li>Figure 3 illustrates a method for abstracting characterizing information descriptive of an address block from an image of the address block.</li><li>Figure 4 illustrates another method for abstracting characterizing information descriptive of an address block from an image of the address block.</li><li>Figure 5 illustrates another method for abstracting characterizing information descriptive of an address block from an image of the address block.</li><li>Figure 6 shows a flow diagram of the operation of a controller comprised in the system of Figure 2 in accordance with one embodiment of the subject invention.</li><li>Figure 7 shows a flow diagram of the operation of the operation of a controller comprised in the system of Figure 2 in accordance with another embodiment of the subject invention.</li></ul>

0016In Figure 2, mailing system 22 includes address printer controller 12, address printer 14, postage meter 16, and indicia printer 20, which are substantially similar to the corresponding prior art elements shown in Figure 1. System 22 differs in including data stores 24 and 26 communicating with controller 12. Data store 24 stores a plurality of characterizing algorithms, as will be described further below, and data store 26 stores at least a print/scan filter which, when applied to the pristine image, generates a filtered image which approximates the transformation of the pristine image by the printing and scanning processes. In other embodiments, data store 26 stores one or more defacing filters which simulate blots, smudges, failure of print elements or scanner sensors, or other, similar occasional events which can not easily be incorporated into said print/scan filter to create one or more defaced images, Together, meter 16, printer 20, form secure postal indicia printing system 22.

0017Three methods for generation of image-based characterizing information, which are believed to provide improved compactness and robustness, have recently been developed by the assignee of the present application and are described below as illustrative of the type of characterizing algorithms which can be used with the subject embodiment. Numerous other algorithms will be apparent to those skilled in the art and particular choices of algorithms to be used form no part of the subject invention, except as may be recited in the claims below. Each of these methods is believed to provide a sufficiently high likelihood of detection to deter "rubber stamp" counterfeiting, particularly by large scale matters, while having a sufficiently low rate of false positives that it will not unduly delay mail processing. It is believed that each of these methods will in general provide characterizing information which can be specified by a bit stream of approximately 6 to 12 bytes.

0018A characterizing algorithm in which the characterizing information comprises measurements of the lengths of the individual words which make up address A, is shown in Figure 3. Address block A is parsed to identify individual words by first identifying line spaces <i>ls</i> by determining the occurrence of large amounts of horizontal white space between blocks of printed text, and then identifying word spaces <i>ws</i> by determining the occurrence of large amounts of vertical white space between blocks of printed text (as shown with respect the first line of address A). Word lengths /1 through /9 are then determined for address A. Preferably, word lengths are taken (measured in pixels) from the edges of word spaces <i>ws</i> (or the address edges) as shown, but can be taken in any convenient manner, such as along the midline of the words.

0019It is believed that using four or fewer bits per word would not be useful in postal applications. Thus, in a preferred embodiment the number of bits used can be selected to encode all words in the address, and two control bits will be sufficient to indicate selection of five to eight bits per word to encode the length of the word. In other embodiments, a fixed number of words in the address, for example the first eight, can be scanned at a fixed number of bits per word; eight in this case, since control bits would not be needed to specify the number of bits per word.

Example

0020An address such as shown in Figures 3 - 5 may produce, depending on the print font selected, etc., the following results using six bits per word: <tables id="tabl0001" num="0001"><table frame="all"><tgroup cols="10"><colspec colnum="1" colname="col1" colwidth="26mm" /><colspec colnum="2" colname="col2" colwidth="11mm" /><colspec colnum="3" colname="col3" colwidth="10mm" /><colspec colnum="4" colname="col4" colwidth="11mm" /><colspec colnum="5" colname="col5" colwidth="10mm" /><colspec colnum="6" colname="col6" colwidth="11mm" /><colspec colnum="7" colname="col7" colwidth="11mm" /><colspec colnum="8" colname="col8" colwidth="11mm" /><colspec colnum="9" colname="col9" colwidth="10mm" /><colspec colnum="10" colname="col10" colwidth="11mm" /><thead><row><entry valign="top">Word#</entry><entry valign="top">1</entry><entry valign="top">2</entry><entry valign="top">3</entry><entry valign="top">4</entry><entry valign="top">5</entry><entry valign="top">6</entry><entry valign="top">7</entry><entry valign="top">8</entry><entry valign="top">9</entry></row></thead><tbody><row><entry>Length(pixels)</entry><entry>173</entry><entry>45</entry><entry>150</entry><entry>60</entry><entry>154</entry><entry>103</entry><entry>168</entry><entry>68</entry><entry>189</entry></row></tbody></tgroup></table></tables> The absolute lengths are then normalized to the range 1 - 63, i.e. 2⁰ - (2⁶ -1), yielding: <tables id="tabl0002" num="0002"><table frame="all"><tgroup cols="10"><colspec colnum="1" colname="col1" colwidth="35mm" /><colspec colnum="2" colname="col2" colwidth="10mm" /><colspec colnum="3" colname="col3" colwidth="10mm" /><colspec colnum="4" colname="col4" colwidth="10mm" /><colspec colnum="5" colname="col5" colwidth="10mm" /><colspec colnum="6" colname="col6" colwidth="10mm" /><colspec colnum="7" colname="col7" colwidth="10mm" /><colspec colnum="8" colname="col8" colwidth="10mm" /><colspec colnum="9" colname="col9" colwidth="10mm" /><colspec colnum="10" colname="col10" colwidth="10mm" /><thead><row><entry valign="top">Word#</entry><entry valign="top">1</entry><entry valign="top">2</entry><entry valign="top">3</entry><entry valign="top">4</entry><entry valign="top">5</entry><entry valign="top">6</entry><entry valign="top">7</entry><entry valign="top">8</entry><entry valign="top">9</entry></row></thead><tbody><row><entry><b>Length(normalized)</b></entry><entry>56</entry><entry>1</entry><entry>46</entry><entry>7</entry><entry>48</entry><entry>26</entry><entry>54</entry><entry>11</entry><entry>63</entry></row></tbody></tgroup></table></tables>

0021Another algorithm in which the characterizing information comprises measurements of the number of "outliers" in each word (or each line) that make up address A, is shown in Figure 4. (By "outliers" herein is meant ascenders or descenders and portions capitals of which project beyond thresholds, which are preferably determined by the upper and lower bounds of lower case letters without ascenders or descenders, such as "a", "c", "e", etc.) Address A is parsed to identify individual words, if necessary, by first identifying line spaces <i>ls</i> by determining the occurrence of large amounts of horizontal white space between blocks of printed text, and then identifying word spaces <i>ws</i> by determining the occurrence of large amounts of vertical white space between blocks of printed text (as shown with respect the first line of address A). Otherwise, only the lines need be identified.

0022Assuming six bits are allocated per word, the number of upwards (+) and downwards (-) outliers per word can be encoded as "xxx/yyy" where x and y are binary digits and xxx is the number of (+) outliers and yyy is the number of (-) outliers.

Example

0023Again taking eight bytes as the space allocated for the address block characterizing information, as shown in Figure 4 with respect to the first address line, (+) outliers 32, in word 1; 34, in word 2; and 36, in word 3 are identified as exceeding threshold 40, and outlier 42, in word 1, is identified as exceeding threshold 44. Since for address block A all of the outliers can be encoded in less than 60 bits, the resulting bit stream is: <tables id="tabl0003" num="0003"><img file="EP1544791B1_D0001.tif" /></tables> where code 1 indicates per word characterization and 111 is an end code. (The 111 end code of course implies that no more than six (+) outliers can be recognized in any word, i.e. 110 means 6 or more.)

0024Another algorithm in which the characterizing information comprises a description of the shape of the address block is shown in Figure 5. The shape is determined by using a conventional "best fit" scanning algorithm which encloses address block A with "best fit" closed curve 50. (It should be understood that various algorithms for generating a best fit curve will generate different curves. These differences do not affect the subject embodiment so long as the same algorithm is used to generate the curve whose description is incorporated into the indicium and to recover the curve from the address block when the indicium is validated.) Preferably, curve 50 is constrained. That is the manner in which a curve can be generated and is limited so that the resulting curve is simplified and can be described with limited information. In Figure 5, curve 50 is formed from linked straight line segments, such as segment 51, which are limited to eight "directions", up (U), down (D), left (L), right I, up-right (UR), up-left (UL), down-right (DR), and down-left (DL); viewed as being generated starting in the upper left corner of address block A and traveling clockwise around address block A. Preferably, the curve 50 also accounts for spaces between characters, words and lines, treating these spaces as equivalent to printed space, so that curve 50 does not become too convoluted and require extensive descriptive information. It is within the skill of a person skilled in the art to provide an algorithm which will generate robust and compact characterizing information, as described above.

0025The characterizing information, i,e., the description of curve 50, can be encoded in a number of ways. In the present example, the characterizing information consists of only the directions, without lengths, of each successive line segment.

Example

0026Encoding line segment directions as: <maths id="math0001" num=""><math display="block"><mi mathvariant="normal">R</mi><mo mathvariant="normal">=</mo><mn mathvariant="normal">000</mn><mo mathvariant="normal">,</mo><mi mathvariant="normal">L</mi><mo mathvariant="normal">=</mo><mn mathvariant="normal">111</mn><mo mathvariant="normal">,</mo><mi mathvariant="normal">U</mi><mo mathvariant="normal">=</mo><mn mathvariant="normal">001</mn><mo mathvariant="normal">,</mo><mi mathvariant="normal">D</mi><mo mathvariant="normal">=</mo><mn mathvariant="normal">110</mn><mo mathvariant="normal">,</mo><mi>UR</mi><mo mathvariant="normal">=</mo><mn mathvariant="normal">010</mn><mo mathvariant="normal">,</mo><mi>DL</mi><mo mathvariant="normal">=</mo><mn mathvariant="normal">101</mn><mo mathvariant="normal">,</mo><mi>DR</mi><mo mathvariant="normal">=</mo><mn mathvariant="normal">011</mn><mo mathvariant="normal">,</mo><mi>UL</mi><mo mathvariant="normal">=</mo><mn mathvariant="normal">100</mn><mo mathvariant="normal">;</mo></math><img file="EP1544791B1_D0002.tif" /></maths> and starting at the upper left of address block A, curve 50 is described by the bit stream: <tables id="tabl0004" num="0004"><img file="EP1544791B1_D0003.tif" /></tables> Thus, curve 50 can be described in nine bytes, including an end code, which can be indicated by reversal (or repetition) of the immediately preceding segment direction. Again, this bit stream is incorporated into the indicium.

0027Programming of a data processor to analyze scan data to perform imaging operations such as identifying lines and words, measuring the dimensions of letters and words or fitting a curve to an image in accordance with predetermined constraints are well known. Such operations are substantially routine in the character and general pattern recognition arts, for example. Techniques for carrying out such operations are also taught in <nplcit id="ncit0001" npl-type="b"><text>Handbook of Pattern Recognition and Image Processing edited by T Young and K-S Fu, Academic Press, 1986</text></nplcit>, and need not be discussed further here for an understanding of the subject embodiment.

0028Bit streams, such as those described above, comprise ordered sequences of values which are typically, though not necessarily, numeric values associated with words in the address block. (Such bit streams are hereinafter sometimes "characterizing information descriptors" or "descriptors", and such values are hereinafter sometimes "characterizations".) As described above, when an indicium is validated, i.e., tied to the mail piece on which it is printed, at a distant postal facility the descriptor generated from the pristine image and incorporated into the indicium is compared with a descriptor recovered from an image scanned from the address block printed on the mail piece. It will be apparent to those skilled in the art that the recovered image will be transformed with respect to the pristine image by the characteristics of the printing and scanning processes, as well as possibly by the occurrence of occasional events such as blots. Thus, it is important that the algorithm used to characterize the address block be robust, that is, that it produce descriptors that match sufficiently when an indicium is valid, and do not match for invalid indicia, despite small differences between the scanned image and the pristine image. It will also be apparent that the robustness of a particular characterizing algorithm can vary for different address blocks. (As a hypothetical example, the above described algorithm based on word length may be less robust for address blocks printed in a small font while algorithms based on the number of outliers, or address block shape may be relatively insensitive to font size.)

0029Figure 6 shows a flow diagram of the operation of controller 12 in accordance with one embodiment of the subject invention. At step 60 controller 12 obtains a pristine digital image, P, of address block A from a conventional source (not shown) such as a data processing system for preparing a bulk mailing. At step 62 controller 12 carries out printing of address block A in a conventional manner. Preferably, this printing process is carried out concurrently with the selection of a characterizing algorithm but, in other embodiments of the subject invention, printing of address block A can be carried out sequentially or by a separate processor.

0030At step 64, controller 12 inputs a print/scan filter which simulates the printing process of printer 14 and the scanning process to be carried out at a remote postal facility from data store 26 and applies it to image P to generate a filtered image, F, which approximates the image which will be scanned from the mail piece at the postal facility. And at step 66 sets index i equal to 1 and variable R equal to 0.

0031At step 70 controller 12 applies the ith characterizing algorithm C_l to images P and F to generate corresponding descriptors C_l(P) and C_i(F); each comprising a sequence of M characterizations, or values, C_l(P)₁ through C₁(P)_M; C_i(F)₁ through C_l(F)_M. Then at step 72, controller 12 compares descriptors C_i(P) and C_i(F) to estimate a robustness value R_l for the ith algorithm C_l, with respect to a particular image P.

0032The comparison at step 72 is carried out using a comparison algorithm associated with characterizing algorithm C_i and which preferably is the same comparison algorithm used at the postal facility to compare the descriptor recovered from the scanned image with the descriptor incorporated into indicium IN. Preferably the comparison is carried out on a characterization by characterization basis, comparing each C_i(P)_l with the corresponding C_l(F)_l to determine if the characterizations match; i.e. if they are "close enough" as defined by the particular comparison algorithm used. (As a hypothetical example, where the characterizations are word lengths they may be considered to "match" if the lengths differ by no more than one or two units; while if the characterizations are the number of outliers in a word a "match" may require exact equality.)

0033In a preferred embodiment, once descriptors C_l(P) and C_l(F) have been compared, an estimate R_i for the robustness of algorithm C_l, with respect to particular image P, is calculated as: <maths id="math0002" num=""><math display="block"><msub><mi mathvariant="normal">R</mi><mi mathvariant="normal">l</mi></msub><mo mathvariant="normal">=</mo><mi>Total no</mi><mn mathvariant="normal">.</mn><mspace width="1em" /><mi>of</mi><mfenced open="[" close="]" separators=""><msub><mi mathvariant="normal">C</mi><mi mathvariant="normal">i</mi></msub><mrow><mo mathvariant="normal">(</mo><mi mathvariant="normal">P</mi><mo>⁢</mo><msub><mrow><mo mathvariant="normal">)</mo></mrow><mi mathvariant="normal">j</mi></msub></mrow><msub><mrow><mspace width="1em" /><mi>matching C</mi></mrow><mi mathvariant="normal">l</mi></msub><mrow><mo mathvariant="normal">(</mo><mi mathvariant="normal">F</mi><mo>⁢</mo><msub><mrow><mo mathvariant="normal">)</mo></mrow><mi mathvariant="normal">j</mi></msub></mrow></mfenced><mo mathvariant="normal">/</mo><mi mathvariant="normal">M</mi><mspace width="1em" /><mfenced separators=""><mi>for j</mi><mo mathvariant="normal">=</mo><mn mathvariant="normal">1</mn><mspace width="1em" /><mi>through M</mi></mfenced><mo mathvariant="normal">;</mo></math><img file="EP1544791B1_D0004.tif" /></maths> where M is the number of characterizations generated by C_i. (Note that since robustness is defined with respect to small changes in the image, in normal use the filters, and the printing and scanning processes, will be such that the descriptors C_l(P) and C_i(F) will have the same number of characterizations. Otherwise an error condition is generated.)

0034Once estimate R_i is determined at step 74 controller 12 determines if R_i is greater than variable R and, if so, at step 78 controller 12 sets R =R_l and index value I = i. Then, or immediately if R_i is not greater than R, at step 80 controller 12 sets i = i+1. At step 82 controller 12 determines if i+1 is greater than N, the number of characterizing algorithms stored. If not, controller 12 returns to step 70 to test the next algorithm. Otherwise, at step 86 controller 12 sends I and descriptor C_l(P) to meter 16 in a conventional manner for incorporation into indicium IN. The postal facility can then recover I to identify C_l and use C_l to validate indicium IN in a conventional manner. In other embodiments, descriptors can be self-identified by their format, or, if a relatively small number of algorithms is used, the facility can sequentially test using all algorithms, with the assumption that only the algorithm actually used to generate the descriptor will give meaningful results; so that index value I need not be included in indicium IN.

0035Figure 7 shows a flow diagram of the operation of controller 12 in accordance with another embodiment of the subject invention. Similarly to the above described embodiment, at step 90, controller 12 obtains pristine digital image, P, of address block A, at step 94 carries out printing of address block A concurrently with the selection of a characterizing algorithm and, at step 96 inputs a print/scan filter.

0036At step 100, controller 12 inputs defacing filters D₁ through D_T (described above) and applies each of these filters to filtered image F to generate defaced images F*D₁ through F*D_T which approximate scanned images of address blocks which have been defaced by occasional events such as blots. At step 102, controller 12 sets index i equal to 1 and variable R equal to 0

0037At step 104 controller 12 applies the ith characterizing algorithm C_l to images P, F and F*D₁ through F*D_r to generate corresponding descriptors C_l(P), C_l(F) and C_i(F*D₁) through C_i(F*D_T); each comprising a sequence of M characterizations, or values, C_l(P)₁ through C_l(P)_M; C_i(F)₁ through C_l(F)_M, etc. Then at step 108, controller 12 compares descriptors C_i(P) with descriptors C_i(F) and C_i(F*D₁) through C₁(F*D_T) to estimate a robustness value R_i for the ith algorithm C_i, with respect to a particular image P.

0038In a preferred embodiment, once descriptors C_i (P) and C_i(F) have been compared an estimate R_i for the robustness of algorithm C_i, with respect to particular image P, is calculated as: <maths id="math0003" num=""><math display="block"><mtable columnalign="left"><mtr><mtd><msub><mi mathvariant="normal">R</mi><mi mathvariant="normal">i</mi></msub><mo mathvariant="normal">=</mo><mi>Total no</mi><mn mathvariant="normal">.</mn><mspace width="1em" /><mi>of</mi><mo mathvariant="normal">:</mo><mrow><mo mathvariant="normal">[</mo><msub><mi mathvariant="normal">C</mi><mi mathvariant="normal">i</mi></msub><mrow><mo mathvariant="normal">(</mo><mi mathvariant="normal">P</mi><mo>⁢</mo><msub><mrow><mo mathvariant="normal">)</mo></mrow><mi mathvariant="normal">j</mi></msub></mrow><msub><mrow><mspace width="1em" /><mi>matching C</mi></mrow><mi mathvariant="normal">i</mi></msub><mrow><mo mathvariant="normal">(</mo><mi mathvariant="normal">F</mi><mo>⁢</mo><msub><mrow><mo mathvariant="normal">)</mo></mrow><mi mathvariant="normal">j</mi></msub></mrow><mspace width="1em" /><mfenced separators=""><mi>for j</mi><mo mathvariant="normal">=</mo><mn mathvariant="normal">1</mn><mspace width="1em" /><mi>through M</mi></mfenced><mo mathvariant="normal">+</mo></mrow></mtd></mtr><mtr><mtd><msub><mi mathvariant="normal">C</mi><mi mathvariant="normal">i</mi></msub><mrow><mo mathvariant="normal">(</mo><mi mathvariant="normal">P</mi><mo>⁢</mo><msub><mrow><mo mathvariant="normal">)</mo></mrow><mi mathvariant="normal">j</mi></msub></mrow><msub><mrow><mspace width="1em" /><mi>matching C</mi></mrow><mi mathvariant="normal">i</mi></msub><mrow><mo mathvariant="normal">(</mo><mi mathvariant="normal">F</mi><mo mathvariant="normal">*</mo><msub><mi mathvariant="normal">D</mi><mi mathvariant="normal">k</mi></msub><mo>⁢</mo><msub><mrow><mo mathvariant="normal">)</mo></mrow><mi mathvariant="normal">j</mi></msub></mrow><mo mathvariant="normal">/</mo><mi mathvariant="normal">M</mi><mspace width="1em" /><mfenced separators=""><mi>for j</mi><mo mathvariant="normal">=</mo><mn mathvariant="normal">1</mn><mspace width="1em" /><mi>through M</mi><mo mathvariant="normal">,</mo><mi mathvariant="normal">k</mi><mo mathvariant="normal">=</mo><mn mathvariant="normal">1</mn><mspace width="1em" /><mi>through T</mi></mfenced><mo mathvariant="normal">]</mo><mo mathvariant="normal">/</mo><mi mathvariant="normal">M</mi><mo>⁢</mo><mfenced separators=""><mi mathvariant="normal">T</mi><mo mathvariant="normal">+</mo><mn mathvariant="normal">1</mn></mfenced><mo mathvariant="normal">;</mo></mtd></mtr></mtable></math><img file="EP1544791B1_D0005.tif" /></maths> where M is the number of characterizations generated by C_i.

0039Again, similar to the embodiment described above, once estimate R_i is determined at step 110 controller 12 determines if R_i is greater than variable R and, if so, at step 112 controller 12 sets R =R_i and index value I = i. Then, or immediately if R_i is not greater than R, at step 114 controller 12 sets i = i+1. At step 118 controller 12 determines if i+1 is greater than N, the number of characterizing algorithms stored. If not controller 12 returns to step 104 to test the next algorithm. Otherwise, at step 120 controller 12 sends I and descriptor C_l(P) to meter 16 in a conventional manner for incorporation into indicium IN. The postal facility can then recover I to identify C_l and use C_l to validate indicium IN in a conventional manner.

0040In other embodiments, whether or not defacing filters are used, descriptor C_l(F) can be incorporated into indicium IN.

0041It is anticipated that other estimates for robustness of characterizing algorithms will be developed as experience with different applications is gained or will be apparent to those skilled in the art. Accordingly it should be understood that, except for particular recitations in the claims below, details of particular estimates used form no part of the subject invention.