Method, apparatus, and program for detecting specific area

The present invention belongs to the technical field of image processing for detecting a specific area that may exist in a face area in an image such as red eyes from an image taken on a photographic film or an image taken by a digital camera. The present invention relates to a specific area detection method and a specific area detection device that enable high-speed detection of red eyes and the like, and a program for executing these.

In recent years, an image recorded on a film is photoelectrically read, the read image is converted into a digital signal, and then various image processing is performed to obtain image data for recording, and the recording light modulated according to the image data is used. A digital photo printer that exposes a photosensitive material and outputs it as a print has been put into practical use. In a digital photo printer, an image captured on a film is photoelectrically read, and the image is used as digital image data to process the image and expose a photosensitive material. Therefore, it is possible to create a print not only from the image taken on the film but also from the image (image data) taken by a digital camera or the like.

In addition, with the spread of inexpensive color printers such as personal computers (PCs), digital cameras, and inkjet printers in recent years, some users take images taken with digital cameras into PCs, perform image processing, and output them with printers. There are many. Furthermore, in recent years, magneto-optical recording media (MO, etc.) and small semiconductor storage media (SmartMedia) that store images taken with digital cameras^TMAnd compact flash^TM, Etc.), magnetic recording media (flexible discs, etc.), optical discs (CDs, CD-Rs, etc.), etc., directly read image data, perform predetermined image processing, and output a print (hard copy). Printers have also been put into practical use.

By the way, in an image including a person such as a portrait, the most important factor that affects the image quality is the finish of the person. Therefore, the red-eye phenomenon in which a person's eyes (pupils) turn red due to the influence of strobe light emission during shooting becomes a serious problem. It is very difficult to correct red-eye with a conventional photo printer that directly exposes from film. However, in the case of digital image processing such as a digital photo printer, red eyes can be corrected by detecting red eyes by image processing (image analysis) and correcting the brightness and saturation of the red eyes region.

When performing such red-eye correction processing, as a method of detecting red-eye from the image, for example, a face is detected from the image by analysis of image data, and then eyes are detected or a red circle is detected from the detected face. A method of detecting the above is exemplified. In addition, various face detection methods used for such red-eye detection have also been proposed.

For example, in Patent Document 1, a candidate region presumed to correspond to a person's face is detected from an image, this candidate region is divided into a predetermined number of small regions, and the frequency of changes in density and brightness for each small region. And the size-related feature amount is obtained, and the pattern showing the relationship between the feature amount of each small area when the area corresponding to the face of the person created in advance is divided into the predetermined number is collated with the feature amount. Discloses a method of evaluating the angle of a face candidate region to improve the accuracy of face detection.

Further, in Patent Document 2, a candidate region presumed to correspond to a person's face is detected from an image, and when the density of this face candidate region is within a predetermined range, it is presumed to be a torso with reference to this face candidate region. Based on the presence or absence of a region where the density difference between the set body region and the face candidate region is less than a predetermined value, or based on the contrast of the density and saturation of the face candidate region and the body candidate region. A method for improving the accuracy of face detection by evaluating the accuracy of the detection result of the face candidate region is disclosed.

Further, in Patent Document 3, a candidate region presumed to correspond to a person's face is detected from the image, and among the detected candidate regions, the degree of overlap is determined for the candidate region that overlaps with other candidate regions in the image. A method for improving the accuracy of face detection is disclosed by evaluating that the region having a higher degree of overlap has a higher probability of being a face region.<patcit num="1"><text>Japanese Unexamined Patent Publication No. 2000-137788</text></patcit><patcit num="2"><text>Japanese Unexamined Patent Publication No. 2000-148980</text></patcit><patcit num="3"><text>Japanese Unexamined Patent Publication No. 2000-149018</text></patcit>

<p> Since such face detection requires accuracy and various analyzes, it is usually used for high-resolution image data such as print output (if it is image data obtained by reading a film, it is so-called fine). It is scan data, and if it is a digital camera, it is necessary to perform it with captured image data), and it takes time to process. Moreover, the orientation of the face in the captured image can basically be four directions depending on the orientation of the camera (horizontal position, vertical position, etc.) at the time of photographing. Here, if the orientation of the face is different, the orientation of the eyes, nose, etc. in the vertical and horizontal directions of the screen is naturally different. Therefore, in order to reliably detect the face, all four directions are supported. It is necessary to perform face detection. In addition, the size (size) of the face in the image also varies depending on the shooting distance and the like, and if the size of the face in the image is different, naturally, the positional relationship (interval) of the eyes, nose, etc. in the image will be different. Therefore, in order to reliably detect a face, it is still necessary to perform face detection corresponding to various face sizes.</p><p> Therefore, the red-eye correction process is a process in which red-eye detection, especially face detection, becomes rate-determining and takes a very long time. For example, in the case of the above-mentioned digital photo printer, a high-quality image without red-eye can be stably produced. Although it can output, it is a major factor that reduces productivity.</p><p> An object of the present invention is to solve the problems of the prior art, and it is possible to detect a specific area that may exist in a face area in an image, such as red eyes and eyelids, at high speed. For example, a specific area detection method capable of stably outputting a high-quality image without red eyes and significantly improving the productivity of a printer, and a specific area detection device that executes this specific area detection method. , And to provide a program to execute these.</p>

<p> In order to achieve the above object, the specific region detection method of the present invention detects a specific region candidate in an image, and then performs face detection in a region including the detected specific region candidate to detect a face. The specific region candidate included in the image is specified as a specific region to be detected, and the detection conditions in at least one of the specific region candidate detection and the face detection are changed between the main region and the non-major region of the image. Provided is a specific area detection method characterized by.</p><p> Further, the specific area detection device of the present invention includes a candidate detection means for detecting a specific area candidate from an image of supplied image data and a face for performing face detection in a region including the specific area candidate detected by the candidate detection means. It has a detection means and a specific means for identifying a specific area candidate included in a region where a face can be detected by the face detection means as a specific area to be detected, and at least of the candidate detection means and the face detection means. One provides a specific region detection device characterized in that the detection conditions of the detection target are changed between the main region and the non-major region of the image.</p><p> Further, the program of the present invention includes a candidate detecting means for detecting a specific region candidate from an image of supplied image data, a face detecting means for performing face detection in a region including the specific region candidate detected by the candidate detecting means, and a face detecting means for performing face detection in the region including the specific region candidate detected. , A specific means for specifying a specific area candidate included in a region where a face can be detected by the face detecting means as a specific area to be detected is executed, and at least one of the candidate detecting means and the face detecting means is used. Provided is a program characterized by changing the detection conditions of a detection target between a main part region and a non-main part region.</p><p> In such a specific region detection method, a specific region processing apparatus, and a program of the present invention, the specific region is preferably red-eye, and the main region is a preset central region of an image. Further, it is preferable to set the region where the specific region candidates are concentrated as the main region region at the time of the face detection.</p>

<p> By having the above configuration, the present invention does not require face detection in a region where the specific region does not exist when detecting a specific region existing in the face region in the image such as red eyes and acne, and the specific region can be used. Even if there is a possibility that it exists, the processing time in the region where the possibility is low can be shortened, and thereby it is possible to detect a specific region in the face region such as red eyes at high speed. Therefore, according to the specific region detection method of the present invention, for example, rapid red-eye correction can be performed by performing red-eye detection at high speed. For example, image data obtained by photoelectrically reading a photographic film or a digital camera can be used. In a photo printer that creates a photographic print from the image data taken in, it is possible to stably output a high-quality print without red eyes while minimizing the decrease in productivity.</p>

Hereinafter, the specific area detection method, the specific area detection device, and the program of the present invention will be described in detail based on the preferred examples shown in the accompanying drawings. The following description will be given by exemplifying the case where red eye is detected as a specific region that may exist in the face region in the image, but the present invention is not limited to this.

FIG. 1A conceptually shows the present invention in a block diagram as an example of a specific area detection method and a red-eye detection device using the specific area detection device. Further, the program of the present invention is a program that executes the processes described below. The red-eye detection device 10 (hereinafter referred to as the detection device 10) shown in FIG. 1 (A) detects red-eye as a specific area from the input image to be processed (the image data) and outputs it to the red-eye correction means 20. It includes a region detecting means 12, a red-eye candidate detecting means 14, a face detecting means 16, and a red-eye identifying means 18. As an example, such an image detection device 10 is configured by using a computer such as a personal computer or a workstation, a DSP (Digital Signal Processor) or the like. The detection device 10 and the red-eye correction means 20 may be integrally configured, or the detection device 10 (or further, the red-eye correction means 20) may be used for color / density correction, gradation correction, and electronic change. It may be incorporated in an image processing device (means) that performs various image processing such as double processing and sharpness processing.

In the detection device 10 of the present invention, the processing target image (hereinafter referred to as a target image) for red-eye detection is not particularly limited as long as it is a color image. For example, an image (image is taken) taken on a photographic film by a camera. It may be an image data (image data) obtained by photoelectrically reading the photographic film) or an image (image data) taken by a digital camera. Further, it goes without saying that the target image may not be the captured image itself, but may be an image (image data) subjected to various image processing as needed.

The target image is first supplied to the region detecting means 12 and the red-eye candidate detecting means 14. The area setting means 12 sets the main part area and the non-main part area in the supplied target image, and sets the setting result (for example, coordinate data indicating the pixel number and the area) to the red-eye candidate detecting means 14 and It is supplied to the face detecting means 16.

In the illustrated example, as an example, the area setting means 12 has a template for area setting as shown in FIGS. 1 (B) and 1 (C), and if necessary, the template (or the target). Enlarge / reduce the image), apply this template to the target image, and set the central area of the image ((B) as the central elliptical area, (C) as the central rectangular area) as the main area. It is set, the other area is set as the non-main part area, and the setting result is sent to the red eye candidate detecting means 14 and the face detecting means 16. It should be noted that instead of enlarging / reducing the template, it is possible to have a template having a plurality of sizes according to the expected size of the target image, or a template having a plurality of sizes and enlarging / reducing may be used in combination.

The method for determining the central region of the image as the main region is not particularly limited, and may be appropriately determined according to the appropriateness required for the detection device 10. For example, the larger the main region, the slower the red-eye detection processing time, but the higher the red-eye detection accuracy when viewed as a whole image. Therefore, depending on the processing time and processing accuracy required for the detection device 10. , A method of appropriately determining the central region of the image as the main region is exemplified. Further, a plurality of templates having different sizes of the image center region as the main region may be prepared and selectable, and / or, in a predetermined template, the image center region as the main region may be arbitrarily selected. It may be configurable.

In the present invention, the method of setting the main part region and the non-main part region is not limited to the method of setting the central region and the peripheral region of the image in this way, and various methods can be used. For example, a method of performing image analysis to set a focused region as a main region and other regions as a non-main region is exemplified. The in-focus region in the image may be extracted by a known method. Further, in general, the region corresponding to the main portion in the image, including the strobe photography, often has higher brightness than the background region. Utilizing this, the high-luminance region exceeding the threshold value in the image may be used as the main region, and the other regions may be designated as the non-main region. Alternatively, the strobe guide number, irradiation area, focal length, and distance measurement information are obtained from various information recorded in the image file or magnetic information recorded on the film in the case of APS, and the area irradiated with the strobe is determined. The main part region may be used, and the other region not irradiated with the strobe may be used as the non-main part region. Further, a region in which the red-eye candidate region detected by the red-eye candidate detecting means 14 is concentrated may be set as a main region in response to face detection described later. This point will be described in detail later.

The red-eye candidate detecting means 14 detects a region that may be red-eye, that is, a red-eye candidate from the target image, and obtains red-eye candidate position information (center coordinate position information), region information, number information, and the like. It is supplied to the face detection means 16 and the red-eye identification means 18 as candidate information. As an example, as shown in Fig. 1 (D), if a person is photographed in a scene with three red lamps in the background and this person has a red-eye phenomenon, the image (scene) corresponding to the red lamp is a. , B, and c, and the regions indicated by d and e corresponding to the red eye are detected as red eye candidates and supplied to the face detecting means 16 and the red eye identifying means 18.

The method for detecting the red-eye candidate is not particularly limited, and various known methods can be used. As an example, the red-eye degree (how red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much red-eye-like color) and circularity (how much are red-eye-like) set in advance from a large number of red-eye image samples are extracted by extracting an area having a red hue and a predetermined number of pixels or more are gathered. A method of detecting a region in which the red-eye degree and the circularity exceed the threshold value as a red-eye candidate having a possibility of red-eye is exemplified by using (round squid).

Here, in the detection device 10 of the illustrated example, the conditions for detecting the red-eye candidate are changed between the main part region and the non-main part region previously set by the region setting means 12. As an example, in the main region, even if it is unlikely to be red-eye, it is detected as a red-eye candidate, and in the non-major region, only the region that is likely to be red-eye is detected as a red-eye candidate. Specifically, as described above, when detecting a region in which the red-eye degree and the circularity exceed the threshold value as a red-eye candidate, the threshold value is lowered in the main part region to detect the red-eye candidate. Alternatively, conversely, the red-eye candidate detection may be performed by increasing the threshold value of the non-major region. Therefore, for example, in the example shown in FIG. 1 (D), the red-eye candidates d and e in the main region are red-eye candidates detected at a lower threshold than the red-eye candidates a to c in the non-major region. In the face detection means 16 described later, face detection is performed only around the red-eye candidate. Therefore, this makes it possible to reduce the face detection processing in the non-main part region having low importance, that is, reduce the amount of face detection processing performed by the face detection means 16, and shorten the processing time for red-eye detection.

The detection result of the red-eye candidate by the red-eye candidate detecting means 14 and the setting result of the main part region and the non-main part region by the region setting means 12 are sent to the face detecting means 16. The face detecting means 16 used the red-eye detection result (for example, the above-mentioned position information) to perform face detection in the vicinity including the red-eye candidate detected by the red-eye candidate detecting means 14, and could detect the face in the area including itself. Information on red-eye candidates, or even face detection results, is supplied to the red-eye identification means 18. For example, in the example shown in FIG. 1 (D), face detection is sequentially performed in a predetermined region including each red-eye candidate corresponding to each of the red-eye candidates a, b, c, d, and e. Therefore, for example, a region surrounded by a dotted line is detected as a face region, and the face detecting means 16 detects information that the red-eye candidates d and e are red-eye candidates included in the face region, or further detects them. The information on the face area is supplied to the red-eye identification means 18.

As described above, face detection is a very time-consuming process, but in conventional red-eye detection, red-eye is present because red-eye is detected in the detected face region after face detection. Face detection is performed even in areas where face detection is not performed, and as a result, face detection takes a very long time. On the other hand, in the present invention, after detecting the red-eye candidate, the face is detected only in the predetermined region including the red-eye candidate, thereby eliminating unnecessary face detection in the region where the red-eye does not exist. In red-eye detection, the time required for face detection can be significantly reduced.

The method of face detection by the face detection means 16 is not particularly limited, and various known methods can be used.

As an example, a method of performing face detection using an average face image so-called face template (hereinafter referred to as a face template) prepared in advance from a large number of face image samples is exemplified. In this method, as an example, the face template (or target image) is set up and down as shown in Fig. 2 (A) according to the orientation of the camera at the time of shooting such as vertical position (vertical shooting) / horizontal position (horizontal shooting). And rotate in the left-right direction (rotate from 0 ° 90 ° 180 ° 270 ° on the image surface) to change the direction of the face, and according to the size (resolution) of the face in the image, Fig. 2 (B) ), The face size of the face template (same as above) is changed (enlargement / reduction = resolution conversion) to create a face template with various face orientations and face size combinations, and a face candidate area in the image. Matching (confirmation of the degree of matching) is performed in sequence, and face detection is performed. Instead of rotating and enlarging / reducing the face template, a rotated face template or an enlarged / reduced face template may be created in advance and used for matching. Further, the face candidate region area detection may be performed by means such as skin color extraction and contour extraction.

In addition, face detection using a learning method is also preferably exemplified. In this method, a large number of face images and non-face images are prepared, feature amounts of each are extracted, and a learning method (for example, Boostong, etc.) selected appropriately from the results is used to make a face. Pre-learning is performed to calculate the function and threshold for separating non-face. When performing face detection, features are extracted from the target image in the same manner as in pre-learning, and the function or threshold obtained in pre-learning is used to determine whether the target image is face or non-face. Perform face detection.

In addition, a method that combines shape recognition by edge (contour) extraction and edge direction extraction, and color extraction such as skin color extraction and black extraction, which are disclosed in JP-A-8-184925 and JP-A-9-138471. Alternatively, each method exemplified as a method for detecting a face candidate other than matching using a face template in Patent Documents 1 to 3 can also be used.

Here, in the present invention, also in the face detecting means 16, face detection is performed under different conditions in the main part region and the non-main part region set by the region setting means 12. For example, in the example of FIG. 1 (D), for the red-eye candidates d and e located in the main region, high-precision face detection without erroneous detection or oversight is performed even if the processing time is long, and the face is not detected. Face detection is performed for red-eye candidates a to c located in the main region so that processing can be performed at high speed. Alternatively, a method is also available in which face detection is performed only in the main region and no face detection is performed in the non-major region.

In this way, by performing high-precision face detection in the main part region and performing face detection at high speed in the non-main part region, or by not performing face detection in the non-main part region, the above-mentioned main part region and the above-mentioned main part region can be obtained. Combined with the effect of red-eye candidate detection in which the conditions are changed in the non-main part region, it is possible to perform highly accurate red-eye detection in the important main part region at a very high speed. In this red-eye detection, it may not be possible to properly detect red-eye in the peripheral region, which is regarded as the non-main region, but since the main subject is usually located in the center of the image, the image quality is improved. It is rarely a problem.

The difference in face detection conditions between the main region and the non-major region is not particularly limited, and various modes can be used. For example, in the non-main part region, a method of increasing the threshold value for determining that the face is not a face (or a method of lowering the threshold value in the main part region) in terms of skin color, circularity, matching degree with the face template, etc. is exemplified. Will be done. According to this method, face extraction can be performed with high accuracy in the main region, and the face detection time in the non-major region can be shortened. Further, in the case of face detection by matching using the above-mentioned face template, face detection corresponding to all face sizes is performed in the main part region, and in the non-main part region, for example, face detection of only the standard face size is performed. A method in which face detection is performed only on a predetermined face size, such as performing face detection on a predetermined face size or larger, is also preferable.

It is also preferable to change the face detection method between the main region and the non-major region. For example, in the main part region, face detection by matching using the face template capable of performing highly accurate face detection and face detection using a learning method are performed, and in the non-main part region, the above-mentioned Japanese Patent Application Laid-Open No. 8- Examples of methods disclosed in 184925 and JP-A-9-138471, which can be processed in a short time, are skin color extraction, shape recognition by edge extraction, face detection by skin color extraction, and the like. Alternatively, in the non-main part region, face detection may be performed only by shape recognition by edge extraction or face detection only by skin color detection.

As described above, the detection result of the red-eye candidate by the red-eye candidate detecting means 14 and the red-eye candidate whose face can be detected by the face detecting means 16 are supplied to the red-eye identifying means 18. The red-eye identification means 18 uses these information to identify red-eye candidates whose faces can be detected in the surroundings as red-eyes, and as red-eye detection results in the target image, the position information of each red-eye, information on the area, and the number of red-eyes. Information and the like are supplied to the red-eye correction means 20.

The red-eye correction means 20 performs image processing of the red-eye region of the target image according to the red-eye detection result supplied from the red-eye identification means 18, and corrects the red-eye of the target image. The method for correcting red-eye is not particularly limited, and various known methods can be used. For example, correction processing that corrects red-eye by controlling the saturation, lightness, hue, etc. of the red-eye region according to the amount of image features of the red-eye and the area around the red-eye (which may include the periphery of the face), or simply the red-eye region An example is a correction process for converting a color to black.

Hereinafter, the present invention will be described in more detail by describing the red-eye detection in detail with reference to the flowchart of FIG.

When the target image is supplied and red-eye detection is started, the region setting means 12 first sets which region of the image is the main region and which region is the non-major region as described above, and detects red-eye candidates. The setting result is supplied to the means 14 and the face detecting means 16.

.. Then, the red-eye candidate detecting means 14 starts detecting the red-eye candidate. In the red-eye candidate detection, according to the area setting by the area setting means 12, red-eye candidate detection is performed at a low threshold value in the main part area as described above, and even an area with a low possibility of red-eye is detected as a red-eye candidate area. In the non-major region, red-eye candidate detection is performed at a high threshold value, and only the region with a high possibility of red-eye is detected as the red-eye candidate region. Assuming that a total of m red-eye candidates can be detected, the red-eye candidate detection means 14 sequentially numbers (numbers) the red-eye candidate regions near the center of the image, and the red-eye candidate detection results are obtained from the face detection means 16 and the face detection means 16. Send to red-eye identification means 18.

In the face detecting means 16 that has received the detection result of the red eye candidate, whether the first (n = 1) red eye candidate (point A) is the main part region or the non-main part region according to the region setting by the previous region setting means 12. As an example, in the case of the main part region, face detection is performed by matching using the face template capable of highly accurate face detection, and in the case of the non-main part region, high-speed processing is performed. Performs face detection by shape recognition by possible skin color extraction and edge extraction. The face detecting means 16 that detects the face at the point A sends information on whether or not the face can be detected at the point A to the red-eye identifying means 18.

The red-eye identification means 18 identifies this red-eye candidate as red-eye when a face can be detected at point A according to the face detection result, and when a face cannot be detected at point A, this red-eye candidate Identify that is not red-eye.

When the identification of whether or not point A is red-eye is completed, point A is set to n + 1, and if A&gt; m, face detection is performed for the next point A (red-eye candidate), and so on. Subsequent face detection for red-eye candidates is performed in sequence, and when A&gt; m is reached, that is, when face detection for all red-eye candidates is completed, red-eye detection is terminated.

In the above example, the red-eye candidate detection means 14 and the face detection means 16 (red-eye candidate detection and face detection) perform red-eye detection and face detection according to the same main region and non-major region. However, the present invention is not limited to this, and the red-eye candidate detecting means 14 and the face detecting means 16 may set a main part region and a non-main part region having different positions and sizes. For example, in the red-eye candidate means 14, red-eye candidate detection is performed using the template shown in FIGS. 1 (B) and 1 (C) with the center of the image as the main region, or the same conditions are applied to the entire region without setting the main region. Detects red-eye candidates with. Next, according to the detection result of the red-eye candidate, as shown in FIG. 4, the face detection means 16 has the peripheral region x including the region where the red-eye candidate r is concentrated as the main region and the rest as the non-major region. You may perform face detection with. At this time, for example, the red-eye candidate detecting means 14 supplies the red-eye candidate detection result to the area setting means 12, and the area setting means 12 includes a circular or elliptical meter including an area including a region where the red-eye candidates are concentrated. Or a rectangular area may be set as the main part area.

Further, the red-eye candidate detecting means 14 detects the red-eye candidate under the same conditions for the entire image (that is, the main part region is not set), and the face detecting means 16 detects the face in the set main part region and the non-main part region. The conditions may be changed, or conversely, the red-eye candidate detection means 14 changes the red-eye candidate detection conditions between the set main region and the non-major region, and the face detection means 16 detects the detected red-eye candidates. Face detection may be performed under the same conditions regardless of the area.

Although the specific region detection method, the specific region detection device, and the program of the present invention have been described in detail above, the present invention is not limited to the above embodiment, and various improvements and improvements can be made without departing from the gist of the present invention. Of course, you may make changes.

For example, the above example is an example in which the detection method of the present invention is used for detecting red eyes, but the present invention is not limited to this, and the present invention is not limited to this, and the present invention is not limited to this. Various things that can exist in the face area in the image, such as wrinkles, are set as specific areas, for example, acne candidates are detected in the image, face detection is performed in the vicinity, and acne candidates for which the face can be detected are identified as acne. You may. As a method for detecting a specific region candidate at this time, for example, a method for detecting a region having a color or shape peculiar to the specific region to be detected from the image is exemplified. Further, as in the case of face detection, a method of matching using an average image (template) of a specific region created in advance from a large number of image samples of a specific region to be detected is also preferable. For example, a method of detecting an eyelid by matching using an average eyelid image, that is, an eyelid template prepared in advance from a large number of eyelid image samples is exemplified.

<figref num="1">(A) is a block diagram conceptually showing an example of using the specific area detection device of the present invention as a red-eye detection device, and (B), (C) and (D) explain the red-eye detection of the present invention. It is a conceptual diagram for doing.</figref><figref num="2">(A) and (B) are conceptual diagrams for explaining a face detection method.</figref><figref num="3">It is a flowchart of an example of red-eye detection in the red-eye detection device shown in FIG.</figref><figref num="4">It is a conceptual diagram for demonstrating another example of red-eye detection by this invention.</figref>

Code description

10 (Red-eye) detection device 12 Area setting means 14 Red-eye candidate detection means 16 Face detection means 18 Red-eye identification means 20 Red-eye correction means