JP5109518B2 - Image processing device - Google Patents

Description

  The present invention relates to a technique for specifying an area where a specific object is shown in an image.

  Conventionally, there exists a technique for determining whether or not a specific object is present in an image. In such a technique, when the image to be examined is dark (for example, 70 or less when the lightness is expressed by a gradation value of 0 to 255), there is a problem that the detection rate of the object is lowered. For this reason, for example, in a certain prior art, the number of detection trials and a detection threshold in the object detection unit are set according to the illuminance of the image capturing environment (Patent Document 1).

JP 2006-209227 A

  However, in the above prior art, it is necessary to perform enormous processing in order to obtain a combination of the number of detection trials and a detection threshold for each illuminance.

  In the technical field of identifying an area where a specific object is shown in an image, in order to solve the above problem, a simple process with a certain level of reliability regardless of the brightness of the image to be detected. Therefore, there is a demand for a technique that can detect a specific object from an image.

  An object of the present invention is to detect a specific object from an image with a certain level of reliability, regardless of the brightness of the image to be detected.

In order to achieve the above object, in one aspect of the present invention, the following processing is performed when specifying a portion where a predetermined object exists in an image.
(A) When a predetermined first condition is satisfied, a reference gradation value included in at least a part of the numerical value range of the first image data is converted into a reference gradation value corresponding to a brighter color. By performing the conversion of 1, the second image data is generated. The reference gradation value is a gradation value relating to the color of the pixel of the image data.
(B) Based on the reference gradation value of the color of the pixel of the first image data, the first portion where the predetermined object exists is specified in the image of the first image data.
(C) When the second image data is generated, a predetermined object exists in the image of the second image data based on the reference tone value of the color of the pixel of the second image data. The second part is specified.
(D) On the basis of the first portion of the image of the first image data and the second portion of the image of the second image data, a predetermined object is present in the image of the first image data. Determine the set of existing parts.

  “Determining based on the first portion of the image of the first image data and the second portion of the image of the second image data” means that the second image data does not exist For the case, it means to determine “based on the first portion of the image of the first image data”.

  If it is set as the above aspects, a specific target object can be detected from the image with a certain level of reliability regardless of the brightness of the image that is the detection target.

  The reference gradation value can be a gradation value representing the brightness of the pixel. In such a case, in many cases, it is possible to specify an object with high accuracy as compared with a case where processing is performed based on a gradation value representing one color component.

The reference gradation value can also be a gradation value representing the intensity of the green color component of the pixel.
Even in such an aspect, for example, an object such as a human face can be specified with higher accuracy than in an aspect in which gradation values of other color components are used.

  It should be noted that at least a part of the numerical range for converting the reference gradation value is a range from the reference gradation value corresponding to the darkest color to 25% of the range of the possible range among the possible range of the reference gradation value. It is preferable to include a range.

  According to such an aspect, the reference gradation value is changed to a brighter reference gradation value for a portion having a reference gradation value in a range from the darkest reference gradation value to 25% in the image. A second part can be identified by conversion. For this reason, even when the object is drawn in a dark color in the image, there is a high possibility that the part where the object exists can be specified as the second part.

  When converting the reference gradation value, it is preferable to perform conversion by multiplying the reference gradation value by a constant.

  In the case where the reference gradation value represents the darker color as the reference gradation value is smaller when the other conditions are the same, the constant is preferably set to a value larger than 1. In such an aspect, by multiplying the gradation value by a constant, the difference in the gradation value of each part in the image is expanded so that the object can be more easily specified in the specification of the second part. can do. On the other hand, when the reference gradation value is the same under other conditions, the constant can be set to a value smaller than 1 when the reference gradation value represents a brighter color as the value of the reference gradation value is smaller.

  When converting the reference gradation value, the conversion can also be performed by gamma-converting the reference gradation value. In such an aspect, compared to an aspect in which the reference gradation value is multiplied by a constant, the difference between the gradation values in a plurality of gradation values having values near the minimum value or the maximum value of the reference gradation value. The possibility of disappearing can be reduced. For this reason, when specifying a 2nd part, a 2nd part can be specified with high precision about the whole image of 2nd image data.

  When the reference gradation value is “a reference gradation value that represents a darker color as the reference gradation value is smaller when other conditions are the same”, the gamma curve is convex upward. Is preferred. On the other hand, when the reference gradation value is “a reference gradation value representing a brighter color as the reference gradation value is smaller when other conditions are the same”, the gamma curve is convex downward. Is preferred.

  When converting the reference gradation value, the conversion can also be performed by changing the reference gradation value by a predetermined amount. According to such an aspect, an object exists in the image by increasing the reference gradation value of the portion having the reference gradation value in the range from the minimum reference gradation value to a predetermined value in the image. The second part can be identified. For this reason, for example, even when an object is drawn with a gradation value in a range near the maximum value or the minimum value that is likely to fail to specify the object in the image, the object is specified. It is possible to specify the second portion by expressing the gradation value in a range other than the range where the error is likely to fail.

  The predetermined amount in the modification is preferably 10% to 40% of the width of the range that the reference gradation value can take. With such an aspect, when specifying a portion where a predetermined object exists, the gradation values near the edges that are likely to be overlooked are generally closer to the median value and are in a range that is less likely to be overlooked. Can be converted to a key value.

  When the reference gradation value is “a gradation value that represents a darker color as the reference gradation value is smaller when other conditions are the same”, the following aspect is preferable. In other words, the first condition preferably includes that the average value of the reference gradation values of the pixels of the first image data is smaller than a predetermined threshold value. According to such an aspect, when the image of the first image data is expressed in a dark color as a whole, the generation of the second image data and the identification of the second portion are performed. Then, when the image of the first image data is expressed in a bright color as a whole, the generation of the second image data and the specification of the second portion are not performed. For this reason, the load of the whole process can be reduced, suppressing the grade which reduces the precision which pinpoints a target object.

  The first condition is included in the first area when the image area of the first image data is divided into a first area and a second area surrounding the first area. It is also preferable that the average value of the reference gradation values of the pixels is smaller than a predetermined threshold value.

  According to such an aspect, when the image in the vicinity of the center where the object is likely to exist among the images of the first image data is expressed in bright colors, the second image data Generation and identification of the second part are not performed. For this reason, the load of the whole process can be reduced, without reducing the precision which pinpoints a target object.

  The first object specifying unit may be a detection module that specifies the first part, and may include a first detection module that is previously learned using sample image data. In addition, the second object specifying unit may be a detection module that specifies the second portion, and may include a second detection module that is previously learned using sample image data.

  Note that the first and second object specifying parts may be a single component or may be different components. Also, the first and second detection modules can be a single component or can be separate components.

  The predetermined object can be a human face.

  When specifying the first portion where the predetermined object exists in the image of the first image data, the first portion is detected using a detection module that has been learned using the sample image data in advance. It is preferable to specify. And when specifying the 2nd part in which the predetermined target object exists in the image of the 2nd image data, the 2nd part can be specified using the detection module.

Furthermore, it is preferable to perform the following processing for the detection module.
(E) First sample image data is prepared.
(F) When a predetermined second condition is satisfied, a gradation value related to a pixel color of the first sample image data and a reference gradation value included in at least a part of the numerical value range is brighter Second sample image data is generated by performing a second conversion that converts the reference gradation value corresponding to the color.
(G) Prior to step (b), the detection module is trained using the first sample image data.
(H) When the second sample image data is generated, the detection module is trained using the second sample image data prior to the step (c).

  According to such an aspect, the detection module can be learned by modifying the gradation value similar to the process performed when the detection module specifies the first and second portions. For this reason, compared with the case where learning is performed using sample image data that is not subjected to any processing, the detection module can efficiently perform learning so that the accuracy in identifying the actual object is increased.

  In the case of preparing a plurality of first sample image data, second sample image data is generated for some first sample image data, and for some other first sample image data, It is also possible not to generate the second sample image data.

  When the reference gradation value is “a gradation value that represents a darker color as the reference gradation value is smaller when other conditions are the same”, the following aspect may be adopted. preferable. In other words, the second condition preferably includes that the average value of the reference gradation values of the pixels of the first sample image data is smaller than a predetermined threshold value. According to such an aspect, when the image of the first sample image data is expressed by the gradation value of the bright color as a whole, the generation and learning of the second sample image data are not performed. . For this reason, the load of the learning process as a whole can be reduced.

  The second condition is that pixels included in the third area when the image area of the first sample image data is divided into a third area and a fourth area surrounding the third area. It is also preferable that the average value of the reference gradation values of the set is smaller than a predetermined threshold value.

  According to this aspect, when the image of the portion near the center where the object is highly likely to be present in the image of the first sample image data is represented by the gradation value of the bright color, The generation and learning of the second sample image data are not performed. For this reason, the load of the entire learning process can be reduced without reducing the accuracy of learning for specifying the target object.

  At least a part of a numerical value range that is a target of gradation value conversion in the generation of the second sample image data is equal to at least a part of a numerical value range that is a target of the gradation value conversion in the generation of the second image data. Is preferred. With such an aspect, it is possible to cause the detection module to perform learning so as to increase the specific accuracy of the second portion using the second sample image data. It should be noted that at least a part of the numerical value range that is a target of gradation value conversion in the generation of the second sample image data is at least a part of a numerical value range that is a target of the gradation value conversion in the generation of the second image data. Can be in different ranges.

  It should be noted that at least a part of the numerical value range that is the target of gradation value conversion when generating the second sample image data is the reference gradation value corresponding to the darkest color among the possible range of the reference gradation value. To a range of up to 25% of the possible range width. According to such an aspect, learning is performed by increasing the reference gradation value of a portion having a reference gradation value in a range from the minimum reference gradation value to 25% in the image of the sample image data. it can. For this reason, even when the object is drawn in a dark color in the image of the sample image data, effective learning can be performed based on the object.

  When generating the second sample image data, the conversion is preferably performed by multiplying the reference gradation value by a constant.

  Further, when generating the second sample image data, it is also preferable to perform conversion by performing gamma conversion on the reference gradation value.

  Further, when generating the second sample image data, it is also preferable to perform the conversion by changing the reference gradation value by a predetermined amount. According to such an aspect, learning is performed by increasing the reference gradation value of the portion having the reference gradation value in the range from the minimum reference gradation value to a predetermined value in the sample image data image. Can do. For this reason, for example, even when the target object is drawn with a gradation value in the range near the maximum value or the minimum value in which the target object is likely to fail in the sample image data image, An object can be represented by a gradation value in a range other than the range in which it is difficult to specify the object, and effective learning can be performed for the object.

  The present invention can be realized in various forms, for example, an image processing method and an image processing apparatus, a print control method and a print control apparatus, and a computer program for realizing the functions of these methods or apparatuses. It can be realized in the form of a recording medium recording the computer program, a data signal including the computer program and embodied in a carrier wave, and the like.

A. First embodiment:
A1. Device configuration:
FIG. 1 is an explanatory diagram showing a schematic configuration of an image processing apparatus according to an embodiment of the present invention. The image processing apparatus includes a personal computer 100 that performs predetermined image processing on image data, a keyboard 120, a mouse 130, and a CD-R / RW drive 140 as apparatuses for inputting information to the personal computer 100, and information. A display 110, a printer 22, and a projector 32 are provided as output devices. In the computer 100, an application program 95 operates under a predetermined operating system. By executing the application program 95, the CPU 102 of the computer 100 realizes various functions.

  When an application program 95 for retouching an image is executed and a user instruction is input from the keyboard 120 or the mouse 130, the CPU 102 stores image data in the memory from the CD-RW in the CD-R / RW drive 140. Read. The CPU 102 performs predetermined image processing on the image data and displays an image on the display 110 via the video driver 91. The CPU 102 can also cause the printer 22 to print the image data that has undergone image processing via the printer driver 96. Further, the CPU 102 can project the image data onto the projector 32 via the driver 98 of the projector 32.

  In this embodiment, the printer driver 96 specifies an area where a human face is supposed to exist in the image of the image data. If such an area exists, correction is performed on the image data so that a human face looks more beautiful before printing. When there is no region where a human face is supposed to exist, another correction is performed on the image data, or no correction is performed on the image data. Thereafter, printing is executed based on the image data.

A2. The principle of face area detection:
In the present embodiment, the module that includes the printer driver 96 executes the process of specifying an area where a human face exists in the image. This module is referred to as a “face detection unit 962”. The face detection unit 962 is learned in advance using sample image data before being mounted in the printer driver 96.

  FIG. 2 is a diagram illustrating a method for identifying an area where a human face is likely to exist in an image of image data. The process described below is executed by the face detection unit 962 of the printer driver 96. Note that an area that is likely to have a human face is referred to as a “face area” in this specification.

  In FIG. 2, an image PI <b> 1 to be examined for the presence or absence of a face region is, for example, an image of 320 pixels × 240 pixels. The image data of the image PI1 has only brightness information for the color of each pixel. When specifying the face area, the image area IDW to be examined in the image PI1 of the image data is specified by the detection window DW having a size equal to or smaller than the image size, and the brightness of each pixel of the image area IDW is specified. Retrieve the data. Thereafter, based on the brightness data of the pixel, it is examined whether or not a pattern that seems to be a human face exists in the image area IDW.

  When the study on one image region is completed, the detection window DW is moved in the image PI1. The movement is performed from left to right (see arrow Ah in FIG. 2). When the detection window DW reaches the right end in the image PI1, it is next placed at the left end in the image PI1 and at a position below the previous position. Thereafter, the detection window DW is similarly moved. Here, the detection window DW is moved by a distance dh larger than the width Wh of the detection window DW in one movement in the left-right direction (see arrow Ah in FIG. 2). Further, the detection window DW is moved by a distance larger than the height Wv of the detection window DW in the vertical movement.

  As a result of analyzing the data extracted by the detection window DW, it is not possible to determine that the image area IDW is “a face area”, but if it is determined that “it may be a face area”, The detection window DW is moved by one pixel in the up, down, left, and right directions (see arrow As in FIG. 2). Then, the data of the image area in the detection window DW is analyzed at each position. When it is determined that the image area in any of the detection windows DW after the movement is “a face area”, the movement (the arrow Ah in FIG. 2) is larger than the width (or height) of the detection window DW. Is resumed.

  On the other hand, the images in the detection window DW after moving up, down, left, and right cannot be determined to be “face areas”, but are determined to be “face areas”. In the case, the following processing is performed. That is, the detection window DW is moved by one pixel from the position determined to have the highest possibility of the face area in the upward, downward, left, and right directions. Thereafter, the same processing is repeated for a predetermined range in the image PI1, and in the state where there is a detection window at a position most likely to be a face region, whether or not a face region finally exists around the image region. Is determined. Thereafter, the movement of a distance larger than the width (or height) of the detection window DW (see arrow Ah in FIG. 2) is resumed.

  FIG. 3 is a diagram showing processing for determining whether or not the image area IDW of the data extracted by the detection window DW is a face area. Whether or not the image area IDW of the data extracted by the detection window DW is a face area is determined through 24 stages (indicated by St1 to St24 in FIG. 3). In the determination of the first stage St1, the determination of the next second stage St2 is performed only when the image area data satisfies a predetermined condition. The same applies to the third to twenty-third stages St3 to St23. The image area IDW that satisfies the conditions in the final 24th stage determination is also determined to be a “face area”.

  For an image that satisfies the conditions up to the ninth stage, it is determined that “the face area cannot be determined but may be a face area”. In such a case, as described above, the detection window DW for one pixel is moved in the vertical and horizontal directions (see arrow As in FIG. 2), and data analysis and detection window movement are repeated. It is.

  FIG. 4 is a diagram showing determination processing in a certain stage. In each stage (see FIG. 3), determination using a rectangular filter is performed. In FIG. 4, rectangular filters F11 and F12 are shown as examples of the rectangular filter. Further, the image of the image area IDW to which the rectangular filter is applied is also displayed so as to overlap with the rectangular filter. In the example of FIG. 4, the image in the image area IDW is assumed to be a face image. As an image of the image area IDW, human face eyes, nose and mouth are shown.

  The rectangular filter has a size of 20 pixels × 20 pixels. Here, to simplify the description, it is assumed that the size of the image area IDW of the data extracted by the detection window DW and the size of the rectangular filter are the same. That is, the image area IDW also has a size of 20 pixels × 20 pixels. The rectangular filter is used for extracting brightness information of pixels included in a part of the image area IDW (shown with hatching in FIG. 4).

  In the determination using the rectangular filter F11, first, the brightness data Y11a of each pixel of the area A11a is extracted from the data of the image area IDW using the rectangular filter F11. The area A11a is a rectangular area that is above the center in the height direction in the image area IDW and has a width equal to the horizontal width of the image area IDW. Similarly, the brightness data Y11b of each pixel in the area A11b is also extracted using the rectangular filter F11. The area A11b is a rectangular area that is below the center in the height direction in the image area IDW and has a width equal to the horizontal width of the image area IDW.

  Note that the area A11a is an area that is estimated to have human eyes when the image area IDW is a face area. The area A11b is an area that is estimated to have a human mouth when the image area IDW is a face area (see the upper left side of FIG. 4).

  The total value of the value obtained by multiplying the sum of the brightness Y11a of each pixel in the region A11a by α11a (α11a is a constant), the value obtained by multiplying the sum of the brightness Y11b of each pixel in the region A11b by α11b (α11b is a constant) is It is determined whether or not it is larger than a predetermined reference value θ11. When the total value is larger than θ11, the value D11y is returned as the value A11 of the determination result using the rectangular filter F11. When the total value is equal to or smaller than θ11, the value D11n is returned as the value A11 of the determination result using the rectangular filter F11 (see the upper right side of FIG. 4). The values D11y and D11n are predetermined constants.

  Similarly, in the determination using the rectangular filter F12, the brightness data Y12a of each pixel of the area A12a is extracted from the data of the image area IDW using the rectangular filter F12. The area A12a is a part of the left half area A11al of the area A11a and includes the center of the area A11al. Also, using the rectangular filter F12, brightness data Y12b of each pixel in the area A12b is extracted from the data in the image area IDW. The area A12b is a part of the right half area A11ar of the area A11a and includes the center of the area A11ar.

  The area A12a is an area that is estimated to have a human right eye when the image area IDW is a face area. Further, the area A12b is an area that is estimated to have a human left eye when the image area IDW is a face area (see the left side in the middle of FIG. 4).

  Then, the sum of the value obtained by multiplying the sum of the brightness Y12a of each pixel in the region A12a by α12a (α12a is a constant), the value obtained by multiplying the sum of the brightness Y12b of each pixel in the region A12b by α12b (α12b is a constant) is It is determined whether or not it is larger than a predetermined reference value θ12. When the total value is larger than θ12, the value D12y is returned as the value A12 of the determination result using the rectangular filter F12. When the total value is equal to or smaller than θ12, the value D12n is returned as the value A12 as a result of determination using the rectangular filter F12 (see the middle right side of FIG. 4). The values D12y and D12n are predetermined constants.

  As described above, in one stage of processing, determination using one or more rectangular filters is performed. Then, the determination result values A11, A12. . . Is determined to be larger than a predetermined reference value Θ1 (see the lower part of FIG. 4). When the total value is larger than Θ1, it is determined that the image area IDW satisfies the determination condition of this stage. When the total value is Θ1 or less, it is determined that the image area IDW does not satisfy the determination condition of this stage. If the image area IDW does not satisfy this stage determination condition, the image area IDW is determined not to be a face area, and the process ends. On the other hand, when the image area IDW satisfies this stage determination condition, the next stage is determined (see FIG. 3).

  The above process is performed using detection windows of a plurality of types. For example, a plurality of types of detection windows having a size of 20 pixels × 20 pixels to a size of 200 pixels × 200 pixels are used for detection of the face area. For this reason, even if faces are drawn in various sizes in the image, they can be detected as face areas. Note that a detection window having an area (number of pixels) of 50% or more of the detection target image PI1 is effective for recognition of a face area when a single face is greatly reflected in the entire image, for example, as in an ID photo. It is.

  In the first embodiment, the size of the rectangular filter is 20 pixels × 20 pixels. For this reason, when a detection window DW having a size other than 20 pixels × 20 pixels is used, the resolution of the image of the data extracted by the detection window DW is converted to 20 pixels × 20 pixels, and the above determination is made. Be targeted.

  The above determination is made based on the brightness of the pixel. In the image represented by the brightness of each pixel, the object is represented by the brightness difference of each pixel. That is, the above determination is performed based on a difference in brightness between pixels in a partial area of the image (see areas A11a, A11b, A12a, and A12b in FIG. 4) and other partial areas. For this reason, there is a possibility that the face area of the person who appears dark overall will not be determined to be a “face area”. For example, as shown in FIG. 2, when a photograph is taken with a flash at night, the flash light reaches the person P1 in the foreground and the flash light does not reach the person P2 in the back. There is. In such a case, as a result of the above processing, the face area Af1 of the person P1 is determined to be “face area”, while the face area Af2 of the person P2 is determined to be “face area”. There is a possibility that it is not determined to be “present” (see FIG. 2).

  On the other hand, there is a possibility that a face area of a person whose light is too strong and the entire face is white will not be determined as a “face area”. For example, when a photograph is taken with a flash at night, the person P2 in the back is moderately illuminated by the light of the flash, and the person P1 in the foreground is exposed to the light of the flash too much and the entire face is white. It may become. In such a case, the face area Af2 of the person P2 is determined to be “face area”, while the face area Af1 of the person P1 may not be determined to be “face area”. There is sex.

  It should be noted that what kind of rectangular filter is used in the above processing is determined in advance prior to the detection of the face region in the learning of the face detection unit 962 performed using the sample image data. Further, reference values such as θ11, θ12, and Θ1 are also determined in learning by the face detection unit 962. On the other hand, coefficients such as α11a and α11b are determined in advance in association with the rectangular filter. That is, in learning, when the rectangular filter used in each stage is determined, the coefficient used is also determined simultaneously.

A3. Face area detection process:
FIG. 5 is a flowchart showing processing in learning by the face detection unit 962. In step S110, a first learning sample image data group is prepared. The first learning sample image data constituting the first learning sample image data group is, for example, image data of 20 pixels × 20 pixels. The first learning sample image data has information on only the brightness represented by the gradation values of 0 to 255 for each pixel. A gradation value 0 represents the darkest lightness, and a gradation value 255 represents the brightest lightness. The first learning sample image data group includes, for example, 10,000 image data in which a face actually exists in the image, and 20,000 image data in which no face exists in the image, for example.

  In step S120, it is determined whether or not the first learning sample image data group has an average brightness of all pixels equal to or less than Ths. If the average brightness of all pixels is equal to or less than Ths (hereinafter referred to as “dark sample data”), the process proceeds to step S130. If dark sample data does not exist in the first learning sample image data group, the process proceeds to step S150. Ths can be set to 70, for example. Note that the first learning sample image data group is preferably prepared in advance so as to include dark sample data in which the average brightness of all pixels is equal to or less than Ths.

  In step S130, a second learning sample image data group is generated based on the dark sample data in the first learning sample image data group. The second learning sample image data constituting the second learning sample image data group is generated by replacing the lightness of each dark sample data pixel of the first learning sample image data group with a lighter brightness.

  In step S140, the face detection unit 962 learns using the second learning sample image data group.

  In step S150, learning by the face detection unit 962 is performed using image data other than dark sample data among the first learning sample image data of the first learning sample image data group.

  In this way, learning is performed using image data in which the average brightness of all pixels is equal to or less than Ths (70) (see steps S120 to S140) and image data in which the average brightness of all pixels is greater than Ths (see step S150). By performing the above, the correct answer rate when detecting the face area in the printer driver 96 can be increased for both the bright image and the dark image.

  FIG. 6 is a flowchart showing processing for determining a region where a face exists in an image of image data. Each step is executed by the printer driver 96.

  First, in step S210, image data for which a face area is specified is read from the CD-R of the CD-R / RW drive 140 (see FIG. 1), and first image data is generated based on the image data. To do. The image data (hereinafter, also referred to as “original image data”) for specifying the face area is, for example, 2560 pixel × 1920 pixel 24-bit color image data. The first image data is image data having an image size of 320 pixels × 240 pixels. The first image data is image data having only brightness information for each pixel.

  The first image data is generated by extracting brightness information of each pixel from the original image data and further performing resolution conversion. The brightness Y of each pixel is obtained by the following equation, for example, when the red, green, and blue tone values are R, G, and B (R, G, B = 0 to 255), for example.

      Y = 0.299R + 0.587G + 0.114B (1)

  In this way, by reducing the resolution of the image and limiting the color information of each pixel to lightness only, the processing load is reduced compared with the aspect of specifying the face area using the target image data as it is. can do. Note that the processing in step S210 is executed by the first image data generation unit 963 as a functional unit of the printer driver 96 (see FIG. 1).

  It is assumed that the image PI0 of the image data that is the target for specifying the face area is a photographic image that was shot with a flash at night. The image PI1 of the first image data is the same image as the image PI0 of the original image data except that the images have different resolutions and are only lightness images (see FIG. 2). In the image PI1, flash light reaches the person P1 in front. For this reason, the person P1 is expressed with lightness and shade of about 70 or more in the image PI1. On the other hand, the flash light does not reach the person P2 in the back. For this reason, the person P2 is expressed in light and shade of lightness of about 70 or less in the image PI1.

  In step S220 of FIG. 6, it is determined whether or not the average brightness AY of all the pixels of the first image data is equal to or less than a predetermined threshold value Thp. When the average brightness AY is equal to or less than Thp, the process proceeds to step S230. If the average brightness AY is greater than Thp, the process proceeds to step S250. Thp can be, for example, a value 70 equal to Ths. By setting Thp to a value equal to Ths, the first image is obtained by utilizing the settings learned by the second learning sample image data (values such as D11y and Θ11 in FIG. 4 and the selected rectangular filter). The face area can be identified from the data image with a high correct answer rate.

  By performing the process of step S220, the processes of steps S230 and S240 are performed only for images that have an average brightness value of 70 or less and that are likely to fail in the face area determination. For this reason, compared with the aspect which processes step S230, S240 about all the images, the burden of the whole process can be eased.

  In step S230, second image data is generated based on the first image data. The second image data is generated by replacing the brightness of the pixels of the first image data with a lighter brightness. Note that the processing of steps S220 and S230 is executed by the second image data generation unit 964 as a functional unit of the printer driver 96 (see FIG. 1).

  In step S240, based on the second image data, one or more face regions that are likely to have a human face in the image are identified. The contents of the process for specifying the face area are as described with reference to FIGS. Note that the process of step S240 is executed by the second face specifying unit 966 as a functional unit of the face detecting unit 962 of the printer driver 96 (see FIG. 1).

  In step S250, one or more face regions are specified based on the first image data. The contents of the process for specifying the face area are as described with reference to FIGS. Note that the processing in step S250 is executed by the first face identification unit 965 as a functional unit of the face detection unit 962 of the printer driver 96 (see FIG. 1).

  In step S260, a set of face areas in the image PI1 represented by the first image data is determined based on the face area specified in step S240 and the face area specified in step S250. Then, based on the face area in the image PI1 represented by the determined first image data, the face area in the image of the image data read from the CD-R of the CD-R / RW drive 140 is determined.

  When determining a set of face areas in the image PI1, for example, for N face areas (N is an integer of 2 or more) that share 75% or more of pixels, one face is selected from the face areas. The area is selected and the other face areas are discarded. By adopting such an aspect, it is possible to prevent the face regions that are mostly overlapped with each other from being determined as a plurality of face regions in the image PI1. The selected face area is, for example, the face area whose center is closest to the center of the area shared by the N face areas. In the present embodiment, the face area is a rectangular area. The center of the rectangular area is the intersection of the diagonal lines of the rectangle.

  In step S260, a set of face areas constituted by the face areas selected in this way is set as a set of face areas in the image PI1. Then, an area in the image of the image data read from the CD-R corresponding to the face area is set as a face area in the image of the image data read from the CD-R. Note that the processing in step S260 is executed by the synthesis unit 967 as a functional unit of the printer driver 96 (see FIG. 1).

  By performing such processing, the face area that can be specified by normal processing can be specified by the processing in step S250. A dark image that is highly likely to fail to specify a face area in normal processing can be specified by the processing in steps S230 and S240. Then, by setting the union (OR set) of these face area sets as the face area in the image of the image data to be examined (see step S260), the face of the bright image area and the image of the dark image area are also determined. Can also be identified as a face region with high probability.

A4. Image data conversion:
Hereinafter, a method for generating the second image data based on the first image data in step S230 of FIG. 6 will be described. The method for generating the second learning sample image data based on the dark sample data in step S130 in FIG. 5 is the same.

  FIG. 7 is a histogram showing a method of generating the second image data based on the first image data in step S230 of FIG. The horizontal axis is the brightness from 0 to 255. The vertical axis represents the frequency of each brightness in the image data. FIG. 7 shows a pixel brightness distribution D1 in the first image data and a pixel brightness distribution D2a in the second image data. The distribution D2a is discrete, but is shown here as a curve for easy understanding.

  The second image data is generated by replacing the brightness of each pixel of the first image data with twice the brightness (see D1 and D2a in FIG. 7). As a result, in the first image data, for example, the brightness of the pixel having the brightness Y1 included in the brightness range R1 of 0 to 70 is the brightness Y2 (Y2 included in the brightness range R2a of 0 to 140. = Y1 × 2).

  Note that the brightness that exceeds 255 as a result of doubling is all replaced with 255. That is, the brightness of the portion expressed by the brightness of 126 or more in the image of the first image data is all replaced with 255.

  By performing such processing, it is possible to convert an image represented by a dark color in the image of the first image data into an image that is brighter and has a large brightness difference (the range R1 in FIG. 7). R2a). As a result, in step S240 of FIG. 6, the face area can be specified with high accuracy by the processing described with reference to FIGS.

  Further, for example, the brightness difference 10 between the lightness 20 pixel and the lightness 30 pixel of the first image data, and the lightness difference 10 between the lightness 120 pixel and the lightness 130 pixel of the first image data, Are doubled to 20 in the first image data after conversion. That is, an equal amount of brightness difference becomes an equal amount of brightness difference after conversion. Therefore, the face area can be accurately identified in the second image data as in the case of the first image data.

  It should be noted that the brightness difference of the portion expressed by the brightness of 126 or more in the image of the first image data is eliminated by the above-described processing for doubling the brightness. Therefore, there is a possibility that the face area cannot be specified for the portion by the processing described with reference to FIGS. However, the processing in steps S230 and S240 in FIG. 6 including the processing in FIG. 7 is performed only when the average brightness of the image data is 70 or less. For this reason, the portion where the brightness difference is eliminated by the processing of FIG. 7 in the target image is not large. Further, for the portion having the brightness of 126 or more, the face area is specified by the face area specifying process (see step S250 in FIG. 6) for the first image data instead of the second image data. For this reason, by performing the process of FIG. 7, there is no possibility that the region where the face is present will fail to be specified as the face region.

  In the first embodiment, the same processing is performed when generating the second learning sample image data based on the dark sample data in step S130 of FIG. That is, the dark sample data is modified and learning is performed in the same process as the process for specifying the face area in the image data (see step S230 in FIG. 6). By setting it as such an aspect, the face detection part 962 can be made to learn so that the precision of the process actually performed may become high.

B. Second embodiment:
The image processing apparatus according to the second embodiment generates a second image data based on the first image data in step S230 of FIG. 6 and a second method based on the dark sample data in step S130 of FIG. The method of generating the second learning sample image data is different from the image processing apparatus of the first embodiment. The other points of the image processing apparatus of the second embodiment are the same as those of the image processing apparatus of the first embodiment.

  FIG. 8 is a diagram showing the content of gamma conversion when generating the second image data based on the first image data in step S230 of FIG. The horizontal axis is the brightness Yi of the pixel before conversion, and the vertical axis is the brightness Yo after conversion of the same pixel. The gamma curve in FIG. 8 is, for example, a gamma curve that replaces lightness 70 with lightness 160. In the second embodiment, in order to generate the second image data, the brightness of each pixel of the first image data is converted according to the gamma curve of FIG.

  Even in such an aspect, an image represented by a dark color in the image of the first image data can be converted into an image that is brighter and has a large brightness difference (see ranges R1 and R2b in FIG. 8). ). As a result, in step S240 of FIG. 6, the face area can be specified with high accuracy by the processing described with reference to FIGS.

  In the second embodiment, the same processing is performed when generating the second learning sample image data based on the dark sample data in step S130 of FIG. That is, the dark sample data is modified and learning is performed in the same process as the process for specifying the face area in the image data (see step S230 in FIG. 6). By setting it as such an aspect, the face detection part 962 can be made to learn so that the precision of the process actually performed may become high.

  Further, in the conversion using the gamma curve, it is less likely that the brightness having different values in the image data before the conversion is converted to the same value (for example, the maximum value 255) as compared with the first embodiment. . For this reason, the face region can be specified for a wide range of brightness.

C. Third embodiment:
The image processing apparatus according to the third embodiment generates a second image data based on the first image data in step S230 of FIG. 6 and a second method based on the dark sample data in step S130 of FIG. The method of generating the second learning sample image data is different from the image processing apparatus of the first embodiment. The other points of the image processing apparatus of the second embodiment are the same as those of the image processing apparatus of the first embodiment.

  FIG. 9 is a histogram showing a method of generating the second image data based on the first image data in step S230 of FIG. The horizontal axis is the brightness from 0 to 255. The vertical axis represents the frequency of each brightness in the image data. FIG. 9 shows a pixel brightness distribution D1 in the first image data and a pixel brightness distribution D2b in the second image data. In the third embodiment, in order to generate the second image data, the brightness of each pixel of the first image data is increased by Δ (Δ is a positive integer, 1 ≦ Δ <255). Δ can be set to 70, for example. In addition, as a result of being increased by Δ, all brightness values exceeding 255 are replaced with 255.

  Even in such an embodiment, an image represented by a dark color in the image of the first image data can be converted into a brighter image (see ranges R1 and R2c in FIG. 9). As a result, in step S240 of FIG. 6, the face area can be specified with high accuracy by the processing described with reference to FIGS.

  There is a higher possibility that an image drawn with pixels having a lightness in the range of lightness 70 or less will fail to specify a face region compared to an image drawn with pixels having a lightness in the other lightness ranges. In the third embodiment, the modification amount Δ is set so that the lightness in the range of lightness 70 or less becomes lightness 70 or more. For this reason, by performing processing on the second image data, the face area can be specified with high accuracy by the processing of FIGS.

  In the third embodiment, the same processing is performed when generating the second learning sample image data based on the dark sample data in step S130 of FIG. That is, the dark sample data is modified and learning is performed in the same process as the process for specifying the face area in the image data (see step S230 in FIG. 6). By setting it as such an aspect, the face detection part 962 can be made to learn so that the precision of the process actually performed may become high.

  In the first and second embodiments, the difference in brightness between pixels changes before and after the brightness change. For example, in the first embodiment, the brightness difference between the lightness 20 pixel and the lightness 30 pixel of the first image data is 10, whereas the corresponding pixels in the second image data The difference in brightness is 20. Therefore, the difference in brightness between the "eye" and the "face background" in the image whose brightness has not been modified, and the difference in brightness between the "eye" and the "face background" in the image whose brightness has been modified, May have different values. That is, there is a possibility that the criterion for determining the brightness difference of the face area is different between a bright image and a dark image. Therefore, when learning is performed by the face detection unit 962 using the first sample image and the second sample image in advance, the learning accuracy may be lowered. The same applies to the second embodiment using a gamma curve.

  On the other hand, in the third embodiment, the brightness of the pixel is modified by adding an integer Δ. That is, the brightness difference between pixels in a dark region is held at the same value before and after the brightness change. For this reason, there is little possibility that the determination criterion of the brightness difference of the face area is different between a bright image and a dark image. Therefore, when the face detection unit 962 learns using the first sample image and the second sample image in advance, the learning accuracy is high.

D. Fourth embodiment:
FIG. 10 is a flowchart showing processing for determining a face area in the fourth embodiment. In the first embodiment, when the average brightness AY of all the pixels of the image data is equal to or less than a predetermined threshold Thp (see step S220 in FIG. 6), the second image data is generated and the face area is specified. (Steps S230 and S240). In the fourth embodiment, as shown in FIG. 10, the second image data is generated without conditioning regarding the average brightness of the image data (see step S230 in FIG. 10). Further, the specification of the face area related to the second image data and the specification of the face area related to the first image data are performed in parallel. Thereafter, a set including both the face area specified in step S240 and the face area specified in step S250 as elements is determined in step S260. The other points of the fourth embodiment are the same as those of the first embodiment.

  The processes of steps S240 and S250 of the fourth embodiment can be executed independently of each other. For this reason, when the personal computer 100 is a computer including a CPU capable of executing multi-threads, it is preferable that the processes in steps S240 and S250 of the fourth embodiment are executed in separate threads. Further, when the personal computer 100 is a computer having a plurality of CPUs, it is preferable that the processes of steps S240 and S250 of the fourth embodiment are executed by separate CPUs. With such an aspect, the entire processing time can be shortened.

E. Variations:
The present invention is not limited to the above-described examples and embodiments, and can be implemented in various modes without departing from the gist thereof. For example, the following modifications are possible.

E1. Modification 1:
In the above embodiment, the image data is converted into image data of 320 pixels × 240 pixels, and the face area is specified (see step S210 in FIG. 6). However, the image data to be processed for specifying the face region is not limited to image data of 320 pixels × 240 pixels, but can be image data having an arbitrary size and number of pixels. However, it is preferable that the image data to be processed for specifying the face area is image data having a certain size. With such an aspect, if one type of face detection module is prepared for image data of that size, it is possible to perform specific processing of the face area.

  The image data is preferably reduced and the number of pixels is reduced before being subjected to a process for specifying a face area. With such an aspect, the processing load can be reduced.

E2. Modification 2:
In the first embodiment, when the average brightness AY of all the pixels of the image data is equal to or less than the predetermined threshold Thp (see step S220 in FIG. 6), the second image data is generated and the face area is specified. (Steps S230 and S240). However, other conditions may be used as conditions for generating image data with a modified tone value (brightness) and specifying a face area.

  For example, an image area of image data is divided into a first area including the center point of the image (for example, an intersection of diagonal lines in the case of a rectangular image) and a second area surrounding the first area. Whether or not to generate face data by generating image data with a modified gradation value according to the average value of a predetermined color or brightness gradation value of a set of pixels included in the first area May be determined. Note that “the second region surrounds the first region” means that any point included in the first region is represented by a point that internally divides the two points included in the second region. Say.

  Also, for example, in the brightness distribution of image data with the horizontal axis as the gradation value and the vertical axis as the frequency, the image data is generated by modifying the gradation value according to the magnitude of the peak gradation value. It may be determined whether or not to specify a region.

  That is, whether or not to generate image data with a modified gradation value and specify a face area can be determined based on the gradation values of the colors of at least some of the pixels of the target image.

  Further, whether or not to generate image data with a modified gradation value and specify a face area can be determined based on whether or not a predetermined process has been performed by the user. That is, when a predetermined process is performed by the user, it is possible to generate image data with a modified gradation value and specify a face area.

  The same applies to the conditions (step S120 in FIG. 5) for generating and learning the sample image data in which the gradation value (lightness) is modified.

E3. Modification 3:
In the above embodiment, in order to generate the second image data, all the brightness values of the first image data are converted (see step S230 in FIG. 6). However, it is also possible to generate the second image data by modifying only the gradation value (brightness) in a partial range of the first image data. In such an aspect, the gradation values in the range from at least the gradation value corresponding to the darkest color to the range of 10% of the range of the gradation value can be taken out of the range that the gradation value can take. It is preferable to carry out. Note that, in the range where the gradation value can be taken, conversion is performed for the gradation value in a range from at least the gradation value corresponding to the darkest color to 20% of the width of the range where the gradation value can be taken, More preferred. Then, conversion is performed for a gradation value in a range from a gradation value corresponding to at least the darkest color to a range of 30% of a range that the gradation value can take, of the range that the gradation value can take, Further preferred.

E4. Modification 4:
In the first embodiment, when the second image data is generated, the brightness of the first image data is doubled. In addition, when the second learning sample image data is generated, the brightness of the dark sample data (first learning sample image data) is doubled. However, the constant for multiplying the gradation value (lightness) by a constant can be a value other than 2. However, the constant is preferably 1.5 to 2.5, and more preferably 1.8 to 2.2.

E5. Modification 5:
In the second embodiment, the gamma conversion replaces the lightness 70 with the lightness 160 when generating the second image data and the second learning sample image data under the condition that the lightness is represented by 0 to 255. Is done. However, the gamma curve used in generating the second image data and the second learning sample image data may have other shapes. However, the gamma curve is preferably a gamma curve in which the lightness 70 is replaced with one of lightness values 100 to 180 under the condition that the lightness is represented by 0 (black) to 255 (white). More preferably, it is a gamma curve in which 70 is replaced with any one of brightness 130-170. The gamma curve is more preferably a gamma curve that replaces the brightness 70 with any one of the brightnesses 140 to 160.

E6. Modification 6:
In the third embodiment, the lightness modification amount Δ is set to 70 under the condition that the lightness is represented by 0 to 255. However, the lightness modification amount Δ can be set to other values such as 60, 50, for example. However, the modification amount Δ of the gradation value (brightness) is preferably 10% to 40%, more preferably 20% to 30%, of the width of the range that the gradation value can take. The gradation value modification amount Δ is more preferably 23% to 27% of the range of the gradation value. When determining whether or not an object exists based on the gradation value of the color of the pixel, the range of gradation values with low accuracy is usually 10 for the entire range that the gradation value can take. % To 40%. Therefore, with the above-described aspect, it is possible to perform the determination by replacing the gradation value included in the range where the determination accuracy is low with a gradation value outside the range.

E7. Modification 7:
In the above-described embodiment and modification, the gradation value of the image when the object is detected is modified and the gradation value of the image when learning is modified by various methods. Whether the modification of the gradation value of the image at the time of detecting the object and the modification of the gradation value of the image at the time of learning are performed in accordance with the same condition (for example, the condition regarding the average brightness of the image) Are preferably determined and performed with the same conversion (eg, multiplying the brightness by a constant). If it is set as such an aspect, efficient learning can be performed so that the precision at the time of detecting a target object may become high.

E8. Modification 8:
In the above embodiment, an area where a human face exists in the image is specified. However, the object specified in the image is not limited to a human face, but may be another object. The processing described in this specification is, for example, processing for identifying various objects such as faces of animals such as dogs and cats, vehicles such as trains and steam locomotives, automobiles, buildings, and plants such as flowers and autumn leaves. Can be applied to.

E9. Modification 9:
In the above-described embodiment, when determining a set of face areas in the image PI1, N face areas that share 75% or more of each other (N is an integer of 2 or more) are included in one face. The area is selected and the other face areas are discarded. However, a plurality of face areas (hereinafter referred to as “candidate face areas”) from which one face area is selected can be determined by other methods. For example, the plurality of candidate face areas may be a plurality of face areas that share 70% or more of pixels. More preferably, the plurality of candidate face areas are a plurality of face areas that share 80% or more of the pixels. More preferably, the plurality of candidate face areas are a plurality of face areas that share 90% or more of each other.

  Furthermore, a plurality of candidate face regions can be determined based on a standard other than the number of pixels to be shared. However, for a plurality of candidate face areas that share a predetermined amount or more of each other, it is preferable to select one candidate face area from among them and determine it as the face area of the image on a predetermined basis.

E10. Modification 10:
In the above embodiment, learning using the first learning sample image data (see step S150 in FIG. 5) and learning using the second learning sample image data with increased brightness (see step S140). Done. Then, the specification of the face area using the first image data (see step S250 in FIG. 6) and the specification of the face area using the second image data with increased brightness (see step S240) are performed. Is called.

  However, the second learning sample image data can also be generated by lowering the color gradation value (lightness) of the first learning sample image data. The second image data can also be generated by lowering the color gradation value (brightness) of the first image data. In such a mode, the target object can be specified even in a range where the entire target object appears white and the difference in gradation values is small.

E11. Modification 11:
In the above-described embodiment, learning using the first learning sample image data (see step S150 in FIG. 5) and learning using the second learning sample image data (for the same face detection unit 962). Step S140). The face area for the image data is also specified using one face detection unit 962.

  However, a plurality of detection modules are prepared, and learning using the first learning sample image data (see step S150 in FIG. 5) is performed on a part of the plurality of detection modules, and the color of the other modules is determined. Learning using the second learning sample image data in which the gradation value (brightness) is modified (see step S140) may be performed.

  After that, for the detection module that has performed learning using the first learning sample image data (see step S150 in FIG. 5), face detection (see step S240 in FIG. 6) for the first image data can be performed. preferable. Then, for the detection module that has performed learning using the second learning sample image data (see step S140), face detection is performed on the second image data whose tone value has been modified (see step S250 in FIG. 6). Is preferably performed.

  In addition, a plurality of types of sample image data groups having different directions and degrees (amounts) of changing the brightness can be provided, and different detection modules can be trained using each group of sample image data. Then, when detecting an object for image data that has undergone a process for modifying the brightness, a sample image in which the brightness is modified in the direction and degree (quantity) corresponding to the direction and degree (quantity) of modifying the brightness. It is preferable to detect an object for the image data with a detection module that has learned data.

  In this aspect, the number of detection modules is the same as the number of groups of images to be learned (in the above embodiment, two groups of the first learning sample image data group and the second learning sample image data group). More preferably.

E12. Modification 12:
In the above embodiment, the face area is specified by the method shown in FIGS. However, the face region can be specified by using various methods such as boosting (for example, AdaBoost), support vector machine, and neural network. However, it is preferable that the face region is specified based on a difference in gradation values relating to the color of each pixel in the image.

E13. Modification 13:
In the above-described embodiment and modification, the face area is identified based on the brightness of the image pixels (FIGS. 2 to 4), the second image data is generated (step S230 in FIGS. 6 and 10), and the second learning is performed. Generation of sample image data (step S130 in FIG. 5) and further determination regarding generation of second image data and second learning sample image data (steps S120 and S220 in FIG. 5) are performed. However, these processes can also be performed based on tone values other than brightness.

  For example, each of the above processes can be performed based on the green gradation value of the image pixel instead of the brightness of the image pixel. Further, each of the above processes can be performed based on the gradation values of red and blue of the pixels of the image instead of the lightness of the pixels of the image. Furthermore, each of the above processes can be performed based on values obtained based on the gradation values of red, green, and blue of the pixels of the image. In other words, each of the above processes can be performed based on the gradation value relating to the color of the pixel of the image.

  Further, the gradation value relating to the color of the pixel used for the processing may be a gradation value relating to the brightness that the image data holds in advance for each pixel. For example, it is possible to use gradation values of luminance and brightness held in JPEG image data and YCrCb color system image data. In addition, the gradation value relating to the color of the pixel used for processing may be a gradation value obtained based on the gradation value of a color component such as RGB held in the image data for each pixel.

E14. Modification 14:
The image processing apparatus may be configured to display a portion where the predetermined object exists after specifying a portion where the predetermined object exists in the image of the original image data. In such an aspect, it is more preferable to display an image of the original image data and to further display a portion where the object exists on the image.

E15. Modification 15:
The image processing apparatus identifies the portion where the predetermined object exists in the image of the original image data, and then based on the size of the portion where the predetermined object exists in the image of the original image data, It is also possible to adopt an aspect in which image processing is performed on the image. In addition, when there are a plurality of portions where the object exists in the image of the original image data, the “size of the portion where the object exists” considered in the image processing is a plurality of portions where the object exists. The size of the largest part can be taken as the size of the part. Further, the “size of the portion where the object exists” considered in the image processing may be the total size of a plurality of portions where the object exists.

E16. Modification 16:
In the above-described embodiment, the printer driver 96 performs processing for specifying the face area, and performs image processing based on the result. However, the process of identifying the object can be performed by other configurations. The process of specifying the object can be executed by, for example, application software 95 executed on the OS of the personal computer 100, or can be executed by the CPUs 104 and 106 provided in the output device such as the printer 22 or the projector 32. . Furthermore, the process of specifying the face area can also be executed by a hardware circuit provided in an output device such as the printer 22 or the projector 32. Furthermore, it can also be executed by a CPU or hardware circuit provided in a digital still camera including an output device such as a liquid crystal display.

  That is, in the above embodiment, a part of the configuration realized by hardware may be replaced by software, and conversely, a part of the configuration realized by software may be replaced by hardware. Good.

  A computer program that realizes such a function is provided in a form recorded on a computer-readable recording medium such as a floppy disk, a CD-ROM, or a DVD. The host computer reads the computer program from the recording medium and transfers it to the internal storage device or the external storage device. Alternatively, the computer program may be supplied from the program supply device to the host computer via a communication path. When realizing the function of the computer program, the computer program stored in the internal storage device is executed by the microprocessor of the host computer. Further, the host computer may directly execute the computer program recorded on the recording medium.

  In this specification, the host computer is a concept including a hardware device and an operation system, and means a hardware device that operates under the control of the operation system. The computer program causes such a host computer to realize the functions of the above-described units. Note that some of the functions described above may be realized by an operation system instead of an application program.

  In the present invention, the “computer-readable recording medium” is not limited to a portable recording medium such as a flexible disk or a CD-ROM, but an internal storage device in a computer such as various RAMs and ROMs, An external storage device fixed to a computer such as a hard disk is also included.

1 is an explanatory diagram illustrating a schematic configuration of an image processing apparatus that is an embodiment of the present invention. The figure which shows the method of pinpointing the area | region where it is likely that a human face exists in the image of image data. The figure which shows the process of determination whether the image area IDW of the data taken out by the detection window DW is a face area. The figure which shows the process of the determination in a certain stage. The flowchart which shows the process in the case of the learning of the face detection part 962. The flowchart which shows the process which determines the area | region where a face exists in the image of image data. The histogram which shows the method of producing | generating 2nd image data based on 1st image data. The figure which shows the content of the gamma conversion at the time of producing | generating 2nd image data based on 1st image data. The histogram which shows the method of producing | generating 2nd image data based on 1st image data. The flowchart which shows the process which determines the face area | region in 4th Example.

Explanation of symbols

DESCRIPTION OF SYMBOLS 22 ... Printer 30 ... Lightness 32 ... Projector 88 ... Host computer 91 ... Video driver 95 ... Application program 96 ... Printer driver 98 ... Projector driver 100 ... Personal computer 102 ... CPU
DESCRIPTION OF SYMBOLS 110 ... Display 120 ... Keyboard 130 ... Mouse 140 ... R / RW drive 962 ... Face detection part A11a, b ... Area where data is extracted using rectangular filter F11 A11al ... Left half area of A11a A11ar ... Right half of A11b Area A12a, b ... Area where data is extracted using the rectangular filter F12 Af1 ... Area where the face exists Af2 ... Area where the face exists Ah ... Arrow indicating the movement of the detection window DW As ... Up, down, left and right D1 ... Pixel lightness distribution in the first image data D2a, b ... Pixel lightness distribution in the second image data D11y, n ... Constants representing results of determination using the rectangular filter F11 D12y , N... Constant indicating the result of determination using the rectangular filter F11 DW... Detection window F11 F12 ... Rectangular filter IDW ... Image area from which data is extracted in the detection window DW P1 ... Person P2 ... Person PI1 ... Image of the first image data R1 ... Lightness range R2a to c ... Range in the image of the first image data Brightness range within R1 St1 to St24 ... 1st to 24th stages Wh ... width of detection window DW Y11a ... brightness of each pixel in area A11a Y11b ... brightness of each pixel in area A11b Y12a ... each of area A12a Pixel brightness Y12b ... Brightness of each pixel in area A12b Yi ... Brightness before conversion Yo ... Brightness after conversion dh ... Moving distance of detection window [alpha] 11a, b ... Constants associated with rectangular filter [alpha] 12a, b ... Rectangular filter Associated constant

Claims (4)

An image processing device for identifying a portion where a human face exists in an image,
When a predetermined first condition is satisfied, a reference gradation value corresponding to a brighter color is changed by changing a reference gradation value included in at least a part of the numerical range of the first image data by a predetermined amount. An image data generation unit that generates second image data by converting into the image data generation unit, wherein the reference gradation value is a gradation value related to a pixel color of the image data;
A first object specifying unit that specifies a first portion where a human face is present in the image of the first image data based on a reference gradation value of a color of a pixel of the first image data When,
A second object specifying unit that specifies a second portion where the human face is present in an image of the second image data based on a reference gradation value of a pixel color of the second image data. And
The human face is present in the image of the first image data based on the first portion of the image of the first image data and the second portion of the image of the second image data. A synthesis unit for determining a part to be performed,
Wherein at least a portion of the numerical range, within the range that the can take reference grayscale value, the reference grayscale value corresponding to the darkest color, seen including a range of up to 25% of the width of the range that can be taken above,
In the determination of the part where the human face exists, the combining unit determines that the first part and the second part share the first part and the first part when the first part and the second part share a predetermined amount or more of the area. An image processing apparatus , wherein one of the second parts is excluded from candidate parts where the human face exists . The apparatus of claim 1 , comprising:
The predetermined amount is 10% to 40% of a width of a range that the reference gradation value can take. The apparatus according to claim 1 or 2, comprising:
The reference gradation value is a gradation value that represents a darker color as the value of the reference gradation value is smaller when other conditions are the same,
The first condition includes an apparatus in which an average value of reference gradation values of pixels of the first image data is smaller than a predetermined threshold value. The apparatus according to claim 1 or 2, comprising:
The reference gradation value is a gradation value that represents a darker color as the value of the reference gradation value is smaller when other conditions are the same,
The first condition is that when the area of the image of the first image data is divided into a first area and a second area surrounding the first area, the first area is defined as the first area. An apparatus including an average value of reference gradation values of included pixels being smaller than a predetermined threshold value.

Images