JP4765523B2 - Image detection apparatus, image detection method, and image detection program - Google Patents

Description

The present invention, images detecting device, a picture image detection method and images detection program.

Conventionally, a specific image included in image data has been detected. In particular, a method for organizing, modifying, and printing photographic data using an indication of the presence or absence of a face image taken by a digital still camera or a scanner, and detecting a face image in the image data has been proposed. In Patent Document 1, a feature vector obtained from a face image as a sample is learned by an SVM (support vector machine), and the feature vector obtained from target image data is in any region with respect to the identification hyperplane. That determines whether or not a face image exists is disclosed. Japanese Patent Application Laid-Open No. 2003-228561 discloses a technique for determining the presence or absence of a face image by preparing a variable template for a face image and matching the variable template with target image data.
According to the former configuration, it is possible to accurately determine the presence / absence of a face image based on a statistical method, and according to the latter configuration, it is possible to determine the presence / absence of a face image even when the facial expression changes. there were.
JP 2005-134966 A JP 2005-149302 A

However, it is inefficient to detect a face image for the entire image data in the above-described technique. That is, only a small part of the image data includes a face image. Therefore, when the face image is detected over the entire area, most of the processing is wasted and the efficiency is low.
The present invention is effectively images detector capable of detecting a specific image from the image data, and an object thereof is to provide a picture image detection method and the images detection program.

Means for Solving the Problems and Effects of the Invention

In order to solve the above-mentioned problem, in the invention according to claim 1, the detecting means sequentially detects a predetermined image for each region in the image data. The areas where the detection means sequentially detects the predetermined image are sequentially set by the area setting means. The region setting means calculates a probability that the predetermined image is detected for the sample image data including the predetermined image with respect to a combination of a relative size and a relative position regarding the sample image data and the predetermined image, and in order from the area specified by the combination of high the relative position and the relative size of probability, and sets an area in the image data. That is, from which region in the image data the predetermined image is detected is set in advance based on the probability that the predetermined image is detected for the sample image data. Since the area is set based on the probability, the order of the selected areas can be set efficiently based on the probability. Therefore, useless processing is prevented, and the detection unit can efficiently detect the predetermined image.

The upper Symbol area setting means for setting the area in the order probabilities above a predetermined image is detected is high in the statistics. Thereby, the detection means can detect the predetermined image in order from the region where the probability that the predetermined image is detected is high, and the predetermined image can be detected at an early stage.

Further, in the above SL probability, the probability that the predetermined image is detected by dividing each relative position of the area in the image data is calculated. The probability that the predetermined image is detected for each position of the region in the image data is statistically determined, and the same region is set in descending order of the probability. That is, the regions at positions where the probability that the predetermined image is detected are high are set in order. When there is a statistical tendency in the position of the predetermined image in the image data, the predetermined image can be detected efficiently.

Further, in the above Symbol statistics, the probability that the predetermined image is detected by dividing each relative size of the area in the image data is calculated. The probability that the predetermined image is detected is calculated for each size of the region in the image data, and the same region is set in descending order of the probability. That is, the regions having a size that increases the probability that the predetermined image is detected are sequentially set. When the size of the predetermined image in the image data has a statistical tendency, the predetermined image can be detected efficiently.

Furthermore, as an example of a suitable detection method for detecting a plurality of the predetermined images, in the invention according to claim 2 , when the first predetermined image is detected in a certain area, the other predetermined images are continued. Will be detected. At that time, the area setting means preferentially sets another area having the same size as the area where the predetermined image has been previously detected. That is, since it is highly likely that a plurality of the predetermined images exist in the same size in the same image data, the other predetermined images in another region having the same size as the region where the predetermined image is detected. Is likely to be detected. Therefore, the predetermined image can be detected efficiently by preferentially detecting from the other area having the same size as the area where the predetermined image is detected.

As another example of a suitable detection method in the case of detecting a plurality of the predetermined images, in the invention according to claim 3 , the region setting means includes the size of the region in which the predetermined image is detected first. The other areas having a small size are preferentially set. That is, in the same image data, there is a high possibility that a plurality of the predetermined images exist in similar sizes. Therefore, in the other region having a size close to the region where the predetermined image is detected, The probability that a predetermined image is detected is high. Therefore, the predetermined image can be detected efficiently by preferentially detecting from the other regions having a size that is small in difference from the size of the region where the predetermined image is detected.

On the other hand, as a preferred example of the method for detecting the predetermined image in the detection means, in the invention according to claim 4 , the detection means compares the feature information extracted from the region with the template of the predetermined image. To do. Various methods can be employed for comparing the feature information. For example, if SVM is used, the presence or absence of the special-purpose image can be accurately determined by performing multivariate analysis of a plurality of feature amounts.

Further, in the invention according to claim 5 , the predetermined image is a human face image. Thereby, a human face image included in the image data can be detected. It can be said that there is a tendency in the position and size where the human face image is arranged in the image data such as a portrait photograph. Therefore, the face image can be efficiently detected by setting the region based on the statistical information.

Of course, the above invention can be realized not only by the apparatus but also by the image detection method as in claim 6 , and by an image detection program for executing processing according to the method as in claim 7 . It can also be realized. In addition, the idea of the invention is that the apparatus, method, and program according to the present invention may be implemented independently, or may be implemented together with other apparatuses, methods, and programs while being incorporated in a certain device. Is not limited to this, and includes various aspects, and can be changed as appropriate. For example, the image detection apparatus of the present invention may be incorporated in an image input device such as a digital still camera. Furthermore, the image detection program of the present invention may be incorporated into a printer driver, a photo retouch program, or an image data management program to function. In the printer driver and photo retouch program, the appropriate image modification method can be switched depending on the presence / absence of a face image, and in the image data management program, image data can be organized depending on the presence / absence of a face image. Good.

  Further, it can be provided as a recording medium on which the program of the present invention is recorded. The recording medium for this program may be a magnetic recording medium, a magneto-optical recording medium, or any recording medium that will be developed in the future. In addition, the duplication stages such as the primary duplication product and the secondary duplication product are equivalent without any question. Further, even when a part is software and a part is realized by hardware, the idea of the invention is not completely different, and a part is stored on a recording medium and is appropriately changed as necessary. It may be in the form of being read. In addition, not all functions are necessarily realized by a single program, but may be realized by a plurality of programs. In this case, what is necessary is just to make each function implement | achieve in a some computer.

Here, embodiments of the present invention will be described in the following order.
(1) Computer configuration:
(2) Flow of face image detection processing:
(3) Modification:
(4) Summary:

(1) Computer configuration:
Figure 1 shows a schematic configuration of a computer as images detecting apparatus of the present invention. In FIG. 1, a computer 10 includes a CPU 11, a RAM 12, an HDD 13, a USB interface (I / F) 14, an input device interface (I / F) 15, and a video interface (I / F) 16 connected by an internal bus 10a. The HDD 13 stores various program data 13a, a plurality of image data 13b, a detection area database (DB) 13c, and template data 13d. The CPU 11 reads out the program data 13a and executes processing based on the program data 13a while using the RAM 12 as a work area. A printer 20 and a digital still camera 30 are connected to the USB interface (I / F) 14, and a mouse 40 and a keyboard 50 are connected to the input device interface 15. Further, a display 60 is connected to the video interface (I / F) 16.

  FIG. 2 shows a software configuration of a program executed on the computer 10. In the figure, an operating system (O / S) P1 is executed, and an application P2 and a printer driver P3 are executed on the O / S P1. The application P2 acquires image data 13b desired by the user from the HDD 13, and accepts a print execution instruction from the user for the image data 13b via the mouse 40 and the keyboard 50. Note that the image data 13b is taken in from the digital still camera 30 to the HDD 13 via the USB I / F 14 in advance. Of course, it may be read from another image input device such as a scanner, or may be read from a network or various recording media. When receiving the print execution instruction, the application P2 transfers the image data 13b to the printer driver P3.

  The printer driver P3 includes an image data acquisition unit P3a, an area setting unit P3b, a detection unit P3c, a determination unit P3d, an image modification unit P3e, and a print data generation unit P3f, and the detection unit P3c further includes a feature information acquisition unit P3c1. It comprises a template comparison unit P3c2. The image data acquisition unit P3a acquires the image data 13b from the application P2. The area setting unit P3b acquires the detection area DB 13c stored in the HDD 13. Then, the area setting unit P3b sets the position and size of the selected area based on the detection area DB 13c.

FIG. 3 shows the contents of the detection area DB 13c. Detection region DB13c selects a number of sample image data including the face image in a random, information indicating exists in any size at any position in the facial image in the sample image data is generated by accumulating. In the figure, the detection area DB 13c describes what size S face image is detected at what frequency A at each position (X, Y). The probability B is calculated by dividing the frequency A by the number of all face images detected from all sample image data. That is, for each combination of position (X, Y) and size S, the probability B that a face image is detected is statistically calculated.

  The position (X, Y) is represented by 20 gradations from 1 to 20, respectively. By dividing the horizontal and vertical pixel coordinates of the center pixel of the face image by the number of horizontal and vertical pixels of the sample image data, the relative position of the center pixel of the face image in the sample image data is calculated. Then, by assigning the calculated relative position to 20 gradations divided into 5% gradation widths, a discrete position (X, Y) that can be statistically processed is obtained. For example, for a face image whose relative position of the center pixel is 0 to 5% from the left end of the sample image data, gradation 0 (representative value 5%) is given, and for a face image whose relative position is 95 to 100% from the left end, gradation is given. 19 (representative value 100%) is given.

  Similarly, the size S of the face image is represented by three gradations from 1 to 3 in ascending order. The relative size of the face image in the sample image data is calculated by dividing the number of pixels of the diameter of the face image by the number of pixels in the horizontal or vertical direction of the sample image data. A discrete size S that can be treated statistically is obtained by assigning the calculated relative size to three gradations divided into predetermined gradation intervals. For each gradation of size S, the representative value of each gradation section is prepared as a relative value with respect to the overall size of the sample image data. For example, representative values of 20%, 50%, and 70% are set for size S of 1 to 3 gradations, respectively.

  Note that the size S may be calculated based on the ratio of the area occupied by the face image to the area of the entire sample image data. Since each of the positions (X, Y) is provided with 20 gradations and the size S is provided with 3 gradations, 1200 combinations of the positions (X, Y) and the sizes S are considered. Based on the size and position of the face image in the sample image data, the face image is attributed to any one of 1200 combinations of position (X, Y) and size S, and the frequency is added by 1 for the corresponding combination. As described above, the detection area DB 13c is statistically obtained with respect to the face image detected from the sample image data, and corresponds to the statistical information in the present invention.

  The region setting unit P3b acquires the detection region DB 13c as described above, and searches the detection region DB 13c for the combination of the position (X, Y) and the size S with the highest probability B. Then, by multiplying the representative value of the relative position (X, Y) and size S by the number of vertical and horizontal pixels of the image data 13b acquired by the image data acquisition unit P3a, it corresponds to the size of the image data 13b. Identify the selection area. In other words, the selected region is set at a position and size with the highest probability B that a face image is detected in the image data 13b. As shown in FIG. 3, since the detection area DB 13c is sorted in descending order of the probability B, the combination of the high position (X, Y) and the size S described in the uppermost column is searched first.

  The feature information acquisition unit P3c1 acquires the size and position of the selected region set by the region setting unit P3b. Then, data belonging to the selected area in the image data 13b is extracted, and predetermined feature information is extracted. Specifically, edge detection or binarization is performed on the data belonging to the selected region in the image data 13b to detect contours that are candidates for each part of the face. A group of feature amounts representing a shape such as a position is extracted as feature information. For example, a parameter indicating at what position and in what size a pixel group expected to be a mouth is extracted as a feature amount. Then, the feature information acquisition unit P3c1 transfers such a group of feature amounts to the template comparison unit P3c2 as feature information.

  The template comparison unit P3c2 acquires feature information from the feature information acquisition unit P3c1 and acquires template data 13d from the HDD 13. The template data 13d is a database in which feature information extracted in advance from a large number of sample face images is accumulated. The template comparison unit P3c2 compares the feature information extracted from the image data 13b by the feature information acquisition unit P3c1 with the feature information of the sample face image stored in the template data 13d. It is determined that a face image has been detected from the selected area of the image data 13b. That is, the template data 13d performs pattern matching between the selected region of the image data 13b and a large number of sample face images by comparing the feature information. If a face image is included in the selected area of the image data 13b, it can be predicted that feature information having a tendency similar to the feature information obtained from the sample face image will be obtained from the selected area. It can be determined that a face image is included in the selected area.

  Since the feature information is composed of a plurality of feature amounts, when the template comparison unit P3c2 compares the feature information, the similarity of the feature information is determined by comprehensively considering each feature amount. Moreover, in order to make this determination with high accuracy, SVM may be used. In SVM, feature information is also prepared for a sample that does not include a face image, and a feature vector group for a sample that does not include a face image in the feature amount space, and a feature vector group of the sample face image that includes the face image. A discrimination hyperplane to be separated is calculated, and a determination is made based on which side the feature vector obtained from the selected region of the image data 13b belongs to the discrimination hyperplane as a boundary. Note that the feature vector is a vector obtained by linearly combining feature amounts, and a weighting coefficient with which SVM has the highest discrimination accuracy is set for each feature amount. In addition, when linear separation cannot be performed in the feature amount space, a kernel trick is used to improve the discrimination accuracy and reduce the processing load.

  As described above, the template comparison unit P3c2 determines whether a face image has been detected for each selected region. When the face image cannot be detected, the area setting unit P3b sets a selection area having the highest probability B that the face image is detected next. The determination unit P3d determines whether or not a face image is detected in any of the selection areas, and determines that no face image is included in the image data 13b if no face image is detected in any of the selection areas. . The image modification unit P3e acquires the image data 13b from which the face image is detected, and performs image modification on the image data 13b. Specifically, the image modification unit P3e performs color tone correction, and changes the color tone correction parameter depending on whether or not a face image is included in the image data 13b. For example, when the face data is included in the image data 13b, parameters are set so that the skin color is corrected to a vivid color, and when the face data is not included in the image data 13b, the correction is not performed. be able to. The print data generation unit P3f acquires the modified image data 13b and converts the image data 13b into a format printable by the printer 20. Specifically, color conversion processing, halftone processing, rasterization processing, and the like are sequentially performed.

(2) Flow of face image detection processing:
FIG. 4 shows the flow of face image detection processing. In the figure, in step S100, the image data acquisition unit P1a acquires the image data 13b instructed to be printed from the application P2. When the image data acquisition unit P1a acquires the image data 13b, the region setting unit P1b acquires the detection region DB 13c in step S110. The region setting unit P1b searches the detection region DB 13c for a combination of a position (X, Y) and a size S with the highest probability B that a face image is detected. Since the detection area DB 13c is sorted in the descending order of the probability B, the combination of the position (X, Y) and the size S having the highest probability B that the face image is detected can be acquired from the top column.

  FIG. 5 is a graph showing the distribution of the center position of the face image, FIG. 6 is a graph showing the probability B for each center position of the face image by contour lines, and FIG. 7 is the size S for each center position of the face image. These graphs are created based on the detection region DB 13c. The relative position (continuous number) in the image data is the vertical axis and the horizontal axis in FIGS. In FIG. 5, it can be seen that the center position of the face image is concentrated near the center of the sample image data. In FIG. 6, it can be seen that the probability B that the center position of the face image exists is higher near the center of the sample image data. When a person is photographed by the digital still camera 30 or the like, it is a normal technique to shoot with the person placed at the center of the photographing region, and the center position of the face image can be concentrated near the center of the sample image data. Because. In FIG. 7, it can be seen that the size S tends to increase toward the center of the sample image data. This is because when the person is placed in the center of the shooting region, the main focus is often placed on the shooting of the person, and there is a high possibility that the main person is zoomed up.

  As described above, there is a statistical tendency in the position and size of the face image in the sample image data, and the region setting unit P1b has the highest probability B that the face image is detected from this statistical information (X , Y) and the size S are searched. In step S130, the selected position is set by converting the searched position (X, Y) and size S into the size of the image data 13b acquired by the image data acquisition unit P1a. That is, since the position (X, Y) and the size S are relative values, the absolute position and size of the selected area are specified based on the size of the image data 13b. For the combination of the position (X, Y) and the size S that have been searched once, a flag indicating that the search has been completed is given, and the search is not performed thereafter.

  FIG. 8 shows the image data 13b and the selected area. In the figure, image data 13b and a selection area C are shown. The selection area C has a rectangular shape, and the height and width thereof are determined by the ratio of the representative value of the size S to the height and width of the image data 13b. The position of the selected area C is set so that the position of the intersection of the diagonal lines of the selected area C matches the coordinates obtained by multiplying the position (X, Y) by the height and width of the image data 13b. ing. However, the position and size of the selection area C need only be set based on the position (X, Y) and the size S, and may be another shape such as an ellipse or a circle.

  In step S140, the detection unit P3c tries to detect a face image from the selection region set in step S130. Specifically, the feature information acquisition unit P3c1 acquires image data belonging to the selection area set for the image data 13b. Then, a group of feature amounts are extracted as feature information while performing predetermined image processing on the image data. The obtained feature information is transferred to the template comparison unit P3c2. The template comparison unit P3c2 acquires the template data 13d from the HDD 13, compares the feature information extracted from the image data 13b by the feature information acquisition unit P3c1 with the feature information of the sample face image stored in the template data 13d. Indicate a similar tendency, it is determined that the face image has been detected from the selected area C of the image data 13b. Note that the absolute size of the selection area C varies depending on the overall size of the image data 13b and the applied size S, so that a face image is detected from the selection area C even when there is similarity to the feature information of the template data 13d. Judge that it was made. Of course, the template data 13d may be prepared for each absolute size of the selection region C, and the feature information may be compared using the template data 13d having an appropriate size.

  In step S150, it is determined whether or not a face image can be detected from the selected area C. If it can be detected, the determination unit P1d determines in step S180 that a face image has been detected from the image data 13b, and detection is performed. The detected position and size of the face image thus output are output, and the face image detection process is terminated. The detection position is the center position of the detected face image, and the size is the center position of the detected face image.

  The fact that the face image has been detected from the image data 13b is output to the image modification unit P3e, and the image modification unit P3e sets parameters for color correction when the image data 13b includes a face image, and performs color correction. Execute. This makes it possible to realize suitable image modification for the image data 13b including the face image, for example, to make the skin color lively. Further, the face image ID, the detection position, and the size of the detected face image may be added to the detection area DB 13c, and the feature information obtained when detecting the face image is added to the template data 13d. You may make it do. By doing so, the number of samples can be gradually increased, and the statistical reliability of the detection region DB 13c can be improved. The image data 13b whose color tone has been corrected by the image modification unit P3e is transferred to the print data generation unit P3f, and print data is generated.

On the other hand, if it is determined in step S150 that the face image could not be detected, it is determined in step S160 whether or not the combination of the unsearched position (X, Y) and size S remains in the detection area DB 13c. . That is, it is determined whether or not there is a position (X, Y) and size S existing in the detection area DB 13c and to which no flag indicating that the search has been performed is added. Then, when the search has been in such have position location (X, Y) and size S remains even one executes step S120 again. However, for the combination of the position (X, Y) and size S searched last time, a flag indicating that the search has been completed is given, so the position (X, Y) and size S of the next highest probability B are assigned. The combination will be searched. Since the detection area DB 13c is sorted in the descending order of the probability B, the searched flag is assigned in order from the upper column to the lower row, and the combination of the position (X, Y) and the size S in order from the upper column to the lower row. Will be searched. Subsequent steps S130 to S150 are performed in the same manner as described above.

  By repeating step S120, the position (X, Y) and the size S are sequentially searched in descending order of the probability B that the face image is detected. In step S130, the selection region C is selected in descending order of the probability B that the face image is detected. Will be set. That is, in step S140, face images are detected in order for the selection region C having a high probability B of detecting a face image. Then, when a face image is detected in any one of the selection areas C, the face image may exist in the normal image data 13b in order to end the process on the assumption that the face image is detected in the image data 13b. It is possible to prevent feature information extraction and template matching from being performed in the selection region C at such a position. Similarly, it is possible to prevent feature information extraction and template matching from being performed in the selection region C having a size such that a face image cannot exist in the normal image data 13b. Therefore, the face image detection process can be performed efficiently.

  If face images are detected for a number of selected areas C, and it is finally determined in step S160 that the combination of the unsearched position (X, Y) and size S does not remain in the detection area DB 13c. Determines that the face image could not be detected from the image data 13b in step S170. Thereby, the image modification unit P3e can recognize that the image data 13b does not include a face image, and can set parameters for color tone correction when the image data 13b does not include a face image. .

  When the face image does not exist in the image data 13b, the face image is detected for combinations of all positions (X, Y) and size S existing in the detection area DB 13c. Therefore, the combination of the position (X, Y) and the size S, which has a certain low probability B, may be invalidated in the detection area DB 13c to increase the processing speed. For example, when the frequency A does not satisfy a significant level with respect to the number of sample image data, the combination of the position (X, Y) and the size S may not be searched in step S120.

  Conversely, a face image may exist in a combination of a position (X, Y) and a size S that does not exist in the detection area DB 13c. Therefore, a position (X, Y) other than the position that exists in the detection area DB 13c and a size S The face image may be detected for the combination. For example, before executing step S170, the position of the selection area C is gradually shifted from the corner of the image data 13b, and the size of the selection area C is sequentially changed at each position to detect a face image. Then, it is possible to detect the face image all over the entire area of the image data 13b. As described above, when all face images are detected for the entire area of the image data 13b, the processing becomes inefficient. In most cases, the face is determined by the combination of the position (X, Y) and the size S existing in the detection area DB 13c. Since the image can be detected, there is no problem.

(3) Modification:
In the above, the image data 13b has been exemplified a printer driver to change the parameters of the color correction depending on whether the face image exists, the images detecting function of the present invention may be incorporated into other programs. For example, it may be incorporated into the images detection function to an image search program for performing image retrieval in order to organize and manage image data photographed by a digital still camera or the like. In an image search program, it is convenient if a search can be performed by specifying not only whether there is a face image in the image data but also how many faces are included.

  FIG. 9 shows a software configuration of an image search program according to this modification. In the figure, an image search program P4 includes a search condition acquisition unit P4e and an image data selection unit P4f in addition to an image data acquisition unit P4a, a region setting unit P4b, a detection unit P4c, and a determination unit P4d similar to the previous embodiment. The output part P4g is comprised. The search condition acquisition unit P4e receives specification of search conditions from the user via the mouse 40 and the keyboard 50. Specifically, it accepts designation of a range for retrieving image data 13b and designation of the number of face images to be retrieved. For example, the user receives a search condition such as searching for image data 13b stored in the C drive that includes four or more face images. This makes it possible to list only the image data 13b containing many face images such as a group photo, which is convenient for designating the number of prints (printing). The image data selection unit P4f sequentially selects the image data 13b belonging to the above-described search range so that the face image is searched for all of the image data 13b belonging to the search range.

  FIG. 10 shows the flow of image data search processing executed in the above configuration. In step S200, the search condition acquisition unit P4e receives specification of the search range and the number of face images from the user via the mouse 40 and the keyboard 50. In step S205, the image data selection unit P4f selects one image data 13b in the specified search range. In step S210, the image data acquisition unit P4a acquires the image data 13b selected by the image data selection unit P4f. Since the process from step S220 to S250 is substantially the same as step S120 to S180 in the previous embodiment, the description thereof is omitted.

  In step S230, a face image is detected in any selected region C, and when a detection result is output in step S245, it is determined in step S255 whether the specified number of face images has been detected. For each piece of image data 13b, each time a detection result is output in step S245, the counter is incremented by one, whereby the number of detected face images can be calculated. In step S255, it is determined whether the number of detected face images has reached the number of face images specified in step S200. If the number of detected face images has reached the number of face images specified in step S200, it is determined that the image data 13b includes at least the number of face images specified by the user. can do. Therefore, in this case, the image data 13b satisfies the user search condition.

  On the other hand, if the number of detected face images does not reach the number of face images specified in step S200, as many face images as specified by the user can be detected from the image data 13b at this time. It can be determined that there is no. Therefore, in this case, the process returns to step S220, and another face image is detected in another selection area C. Then, steps S220 to S245 are repeated until the number of face images designated by the user is detected. However, when returning from step S255 to step S220, step S260 is performed. In step S260, the detection area DB 13c is sorted so that the size of the face image detected immediately before is given priority.

  FIG. 11 shows the detection area DB 13c sorted in step S260. This figure shows the sorting result when the size S (S = 1) is selected in the detection of the face image detected immediately before. In the upper column of the detection area DB 13c, combinations of the size S and the position (X, Y) that are the size S (S = 1) are listed in the upper column, and the probability B in the size S (S = 1). Are sorted in descending order. In step S220, the search area DB 13c is searched from the upper column for which the searched flag is not assigned. Therefore, in the order of increasing the probability B in the size S (S = 1), the size S and The combination of the position (X, Y) will be searched. That is, in step S220, the search is performed with the probability B as the highest priority in the first stage, but once the face image is detected, the combination of the size S and the position (X, Y) of the same size S as the face image is detected. Will be preferentially searched. For items other than size S (S = 1), the probability B is sorted with the highest priority as usual.

  When a plurality of face images are included in the same image data 13b, it can be estimated that the plurality of face images have similar sizes. For example, in a group photo, each person is usually aligned so that the distance from the camera is the same, and the size of the face image to be taken can be considered to be almost the same. Therefore, a face image is highly likely to be detected in the selected area C set by the combination of the size S and the position (X, Y) searched in the sorted image data 13b, and the number of faces specified earlier. An image can be detected. Further, not only the same size S applied to the detection of the immediately preceding face image, but also a size S having a small difference from the size S may be searched preferentially.

  FIG. 12 shows the detection area DB 13c that is sorted so as to be preferentially searched for those having a small difference from the size S applied to the detection of the immediately preceding face image. In the figure, S = 1 is sorted in the upper column, then S = 2 is sorted in the middle column, and finally S = 3 is sorted in the lower column. Since the size S applied to the detection of the immediately preceding face image is S = 1, S = 1 with a difference of 0 is searched first, then S = 2 with a difference of 1 is searched, and finally the difference is S = 3 which is 2 is searched. Since face images having extremely different sizes are unlikely to be included in the same image data 13b, a plurality of face images can be efficiently detected by performing such sorting.

  In the image search program as in this modification, the search may be performed using not only the number of face images but also the characteristics of the face images and people as keys. For example, when detecting a face image in step S230, the feature of the detected face image may be determined. Specifically, characteristics such as gender and age of the detected face image may be determined, and the characteristics may be output and compared with the search condition. Further, when detecting the face image, the face image is identified based on the feature of the detected face image, and the detected face image is that of the person specified as the search condition. It may be determined whether or not.

  In this modification, when no face image is detected in any of the selection areas C (No in S240) and when a specified number of face images are detected (Yes in S255), step S265 is performed. It is determined whether face detection has been completed for all image data 13b within the search range specified in. When face detection has not been completed for all the image data 13b within the search range, the image data selection unit P4f selects the next image data 13b and detects a face image for the image data 13b. In this way, face images can be detected for all image data 13b within the search range. When face detection is completed for all the image data 13b within the search range, the output unit P4g outputs the search list to the display 60 in step S270. The search list displays a list of file names and the like of the image data 13b determined to contain the number of face images specified by the user.

In the above description, the face detection function according to the present invention is executed on a computer. However, the present invention may be incorporated in another apparatus. For example, the present invention may be applied to a digital still camera, and the captured image data may be searched or corrected based on the presence or number of face images. Furthermore, a face image may be specified at the time of shooting, and the focus may be focused on the specified face image. According to the present invention, since a face image can be detected at high speed, it is possible to provide a digital still camera excellent in responsiveness, which can prevent focusing from taking time. Moreover, although what detected the face image as a predetermined image was illustrated, you may detect another image as a predetermined image. That is, if template data 13d for other images is prepared, matching can be performed in the same manner as a face image.

(4) Summary:
In the present invention, combinations of position (X, Y) and size S are sequentially searched in step S120 in descending order of the probability B that a face image is detected in the detection area DB 13c, and the probability that a face image is detected in step S130. The selection area C is set in order of B being higher. As a result, in step S140, the face images are detected in order for the selection region C having a high probability B that the face image is detected. Then, when a face image is detected in any one of the selection areas C, the face image may exist in the normal image data 13b in order to end the process on the assumption that the face image is detected in the image data 13b. It is possible to prevent feature information extraction and template matching from being performed in the selection region C at such a position.

It is a hardware block diagram of a computer. It is a software block diagram of a computer. It is a table | surface which shows detection area DB. It is a flowchart which shows the flow of a face image detection process. It is a graph showing distribution of the center position of a face image. It is the graph which represented the probability for every center position of a face image by the contour line. It is a graph showing the size for every center position of a face image. It is a figure which shows image data and a selection area. It is a software block diagram of the computer concerning a modification. It is a flowchart which shows the flow of the face image detection process concerning a modification. It is a table | surface which shows detection area DB concerning a modification. It is a table | surface which shows detection area DB concerning another modification.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 ... Computer (face image detection apparatus), 10a ... Bus, 11 ... CPU, 12 ... RAM, 13 ... HDD, 13a ... Program data, 13b ... Image data, 13c ... Detection area DB, 13d ... Template data, 14 ... USBI / F, 15 ... Input device I / F, 16 ... Video I / F, 20 ... Printer, 30 ... Digital still camera, 40 ... Mouse, 50 ... Keyboard, 60 ... Display, P1 ... O / S, P2 ... Application, P3: Printer driver, P3a, P4a ... Image data acquisition unit, P3b, P4b ... Area setting unit, P3c, P4c ... Detection unit, P3c1, P4c1 ... Feature information acquisition unit, P3c2, P4c2 ... Template comparison unit, P3d, P4d ... Judgment unit, P3e ... Image modification unit, P3f ... Print data generation unit, P4 ... Image management program P4E ... search condition obtaining unit, P4f ... image data selection section, P4g ... output unit

Claims (7)

An image detection device for detecting a predetermined image included in image data,
For sample image data including the predetermined image to calculate the probability of detecting the predetermined image on the combination of the relative sizes and the relative positions and to a corresponding sample image data and the predetermined image, high the probability the relative position and the Region setting means for setting regions in the image data in order from the region specified by the combination with the relative size;
Detecting means for detecting the predetermined image for each region;
An image detection apparatus comprising: In the case of detecting a plurality of the predetermined images, when the detecting means detects the predetermined images,
The image detection apparatus according to claim 1, wherein the area setting unit preferentially sets another area having the same size as the area where the predetermined image is detected. In the case of detecting a plurality of the predetermined images, when the detecting means detects the predetermined images,
The image detection apparatus according to claim 1, wherein the area setting unit preferentially sets the other area having a small difference from the size of the area where the predetermined image is detected. The detecting means is
The presence / absence of the predetermined image in the region is determined by comparing the feature information extracted from the region with the template of the predetermined image. Image detection device.   The image detection apparatus according to claim 1, wherein the predetermined image is a human face image. An image detection method for detecting a predetermined image included in image data,
The region setting means calculates the probability that the predetermined image is detected for the sample image data including the predetermined image with respect to the combination of the relative size and the relative position regarding the sample image data and the predetermined image, and the probability is high. in order from the area specified by the combination of the relative position and the relative size, an area setting step of setting an area in the image data,
A detecting step for detecting the predetermined image for each of the regions;
An image detection method comprising: An image detection program for causing a computer to execute a function of detecting a predetermined image included in image data,
For sample image data including the predetermined image to calculate the probability of detecting the predetermined image on the combination of the relative sizes and the relative positions and to a corresponding sample image data and the predetermined image, high the probability the relative position and the A function for setting the region in the image data in order from the region specified by the combination with the relative size;
A function of detecting the predetermined image for each of the regions;
Detection program for causing computer to execute

Images