JP2012133573A - Image recognition device, image recognition method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image recognition device capable of accurately follow a target in an image.SOLUTION: An image recognition device specifies a target region from an input image, and follows the target region by calculating an attention point representing a position of the specified target region. When a region including the attention point is different from the target region, the image recognition device corrects the position of the attention point to an internal position of the target region, and follows the target region based on the corrected attention point so as to accurately follow the target even when detecting, for example, a doughnut shaped target.

Description

  The present invention relates to an image recognition apparatus, an image recognition method, and a program, and more particularly to a technique suitable for use in tracking a specific subject.

  2. Description of the Related Art Conventionally, a subject detection function is known in which an image is divided into regions having the same feature amount such as the same color and luminance from the color / luminance information in the image and the subject region is automatically detected from the divided image. A subject tracking function is also known in which a point in a moving image is designated by a user operation from a touch panel or the like, and the area is tracked using color / luminance information of the designated area. Furthermore, a method is also known in which these are combined to specify one point in the subject area detected by the subject detection function and track the subject using the subject tracking function.

  For example, Patent Document 1 discloses a method for tracking a specified object. Specifically, a detection area is set for the tracking point on the moving object based on the direction specified by the user, and the position of the center of gravity is estimated as the tracking point. Thereby, the user can track the object by specifying the tracking point with high accuracy for the moving object.

JP 2007-272731 A

  However, in the conventional technique disclosed in Patent Document 1 described above, when the subject has a donut shape or the like, the center of gravity position may be a non-subject region. There is a possibility that.

  The present invention has been made in view of the above-described problems, and an object thereof is to enable accurate tracking of a subject in an image.

  The image recognition apparatus of the present invention includes a specifying unit that specifies a subject area from an input image, a calculating unit that calculates a point of interest that represents the position of the subject area specified by the specifying unit, and a calculation unit that calculates the target area. Tracking means for tracking the subject area based on the focused point of interest, and the calculating means determines the position of the focused point when the area including the focused point is a different area from the subject area. The correction is made inside the subject area, and the tracking means tracks the subject area based on the attention point of the corrected position.

  According to the present invention, a subject can be accurately tracked even if the center of gravity position of the subject detection region is a region that is not a subject.

It is a block diagram which shows the function structural example of the image recognition apparatus which concerns on embodiment. 10 is a flowchart illustrating an example of a procedure for switching between subject detection and subject tracking. It is a flowchart which shows an example of the process sequence which detects a to-be-photographed area | region. It is a figure explaining the procedure which detects a to-be-photographed area | region by area | region division. It is a flowchart which shows an example of the process sequence which tracks a to-be-photographed object. It is a flowchart which shows an example of the process sequence which calculates and correct | amends an attention point. It is a figure explaining the procedure which correct | amends an attention point.

<Configuration of imaging device>
Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the present embodiment, a case where a digital camera is used as an example of the imaging apparatus according to the embodiment will be described.
FIG. 1 is a block diagram illustrating a functional configuration example of an image recognition apparatus 100 that performs subject detection and tracking in the digital camera according to the present embodiment.
In FIG. 1, an image signal input unit 101 inputs a digital image signal. For example, in an imaging device, incident light is received through an optical unit (not shown) composed of a lens or the like, and a charge corresponding to the amount of light is output by a CCD unit (not shown). A digital image signal obtained by performing sampling, gain adjustment, A / D conversion, and the like is input to the analog image signal output from the CCD unit by an A / D conversion unit (not shown). Furthermore, various image processing is performed on the digital image signal input from the image signal input unit 101 via an image processing unit (not shown), and a processed digital image signal is output. For example, a digital image signal is converted into a YUV image signal and output.

  The subject area specifying unit 102 specifies a subject area in the image. Here, the subject is, for example, a human face, an animal, a building, or the like. As a method for specifying a subject area, there is subject detection that uses color information, luminance information, contrast information, and the like in an image. In addition, as another method for specifying the subject area, designation by a user operation using a touch panel (not shown) or the like disposed on the display surface of the imaging apparatus may be used.

  The subject area attention point calculation unit 103 calculates the attention point from the subject area specified by the subject area specification unit 102. The attention point is, for example, the center of gravity of the subject area. The subject area attention point correction unit 104 corrects the position of the attention point calculated by the subject area attention point calculation unit 103. Whether or not to correct is determined by comparing, for example, the color information of the subject region and the color information around the point of interest, and whether or not they are different. Then, correction is performed when the color information around the subject region and the point of interest is different. As an example of the correction method, the direction in which the attention point is moved is determined from the top, bottom, left, and right in the image, and is moved to the position of the attention point in an area that matches the color information of the subject area.

  The subject tracking unit 105 performs subject tracking based on the position of the target point corrected and determined by the subject region target point correcting unit 104. As a method of tracking the subject, for example, when the tracking of the subject is started, the distribution of the color information and luminance information around the attention point is held, and tracking by pattern matching is performed, so that the attention point with respect to the movement of the subject Keep updating the position of.

  The subject information display unit 106 performs display control to display information indicating the subject at the point of interest position in the image output from the subject tracking unit 105 on a display unit (not shown). The information indicating the subject is assumed to display a rectangular frame around the point of interest, for example. With the above configuration, it is possible to set a point of interest to be used when subject tracking is performed according to the shape of the subject region, and to accurately track the subject.

<Switching between subject detection and subject tracking>
Next, a method for switching between subject detection and subject tracking will be described. FIG. 2 is a flowchart illustrating an example of a processing procedure for switching between subject detection and subject tracking.

  First, in step S201, the subject tracking unit 105 determines whether subject tracking can be executed. When a subject area is detected by subject detection described later, it is considered that subject tracking can be executed. Further, when the process of step S201 is performed for the first time, it is considered that subject tracking cannot be performed because the tracking area has not been determined. As a result of this determination, if subject tracking can be performed, the process proceeds to step S203, and if subject tracking cannot be performed, the process proceeds to step S202.

  In step S202, the subject area specifying unit 102 performs subject detection and detects a subject area in the image. In step S204, it is determined whether or not a subject area has been detected as a result of subject detection executed in step S202. Whether or not the subject area has been detected is determined by, for example, calculating an evaluation value representing the likelihood of the subject when executing subject detection and determining whether or not the evaluation value is equal to or greater than a threshold value. If the subject area is detected as a result of this determination, the process proceeds to step S206, and if the subject area is not detected, the process proceeds to step S207.

  On the other hand, in step S203, the subject tracking unit 105 tracks the subject by tracking the subject area detected by subject detection. In step S205, it is determined whether the subject tracking is successful as a result of the subject tracking executed in step S203. The success or failure is determined, for example, by calculating an evaluation value of the tracking area according to the subject tracking execution result and whether the value is equal to or greater than a threshold value. If the subject tracking is successful, the process proceeds to step S208. If the subject tracking is not successful, the process proceeds to step S207.

  In step S206, the subject tracking unit 105 considers that subject tracking can be performed because the subject region has been detected, and proceeds to step S209 to calculate the attention point. In step S209, the subject area attention point calculation unit 103 calculates the attention point used when subject tracking is performed from the subject area. A detailed description of this attention point will be given later.

  On the other hand, in step S207, the subject tracking unit 105 considers that subject tracking cannot be performed because subject detection or subject tracking has failed. In step S208, the subject tracking unit 105 considers that subject tracking can be executed because the subject tracking is successful.

  As described above, by repeating the processing shown in FIG. 2, it is possible to switch between subject detection and subject tracking. In this embodiment, an example in which either one of subject detection and subject tracking is executed has been described. However, both subject detection and subject tracking may be continuously executed. For example, when a new subject area is detected by subject detection during subject tracking, the subject in the newly detected subject area is tracked according to the evaluation values of subject detection and subject tracking.

<Execution of subject detection>
Next, a method for detecting the subject area from the image in step S202 of FIG. 2 will be described. FIG. 3 is a flowchart illustrating an example of a processing procedure for detecting a subject area by the subject area specifying unit 102.
First, in step S301, the image is divided into regions having different feature amounts based on the feature amounts such as color and luminance in the image. An example of a method of dividing into a plurality of regions will be described with reference to FIG.

  For example, in order to divide the image shown in FIG. 4A into regions having the same feature amount such as the same color and luminance, first, the image is divided into a plurality of blocks. As a block unit, for example, as shown in FIG. 4, the image is divided into 16 × 12 blocks. Then, color information and luminance information between adjacent blocks are compared to determine whether or not those blocks have the same characteristics. Whether adjacent blocks have the same characteristics is determined based on the magnitude of the difference in luminance value or hue value between the blocks. Alternatively, create a histogram of luminance values and hue values for the entire image, set the range of values considered to be the same luminance and the same color from the shape of the peaks and valleys of the histogram, and set the luminance values within the same range between adjacent blocks. Judgment is made based on whether or not it has a hue value. By such area division, it becomes possible to divide into areas having the same feature amount as shown in FIG.

  Next, in step S302, a background area to be regarded as a background is determined for each area in the image divided in step S301, and the determined background area is removed from the subject area candidates. As an example of a method for determining whether or not each area is a background area, an evaluation value (hereinafter referred to as a first evaluation value) of background likelihood is calculated.

  Here, it is considered that there is a high possibility that the subject exists near the center in the image, and the first evaluation value is determined, for example, by the ratio of the number of blocks in contact with the screen edge to the area of the region. In a region where the number of blocks in contact with the screen edge is large, the ratio of the number of blocks in contact with the screen edge to the area of the region is large. For this reason, the first evaluation value is assumed to be high because the possibility of the subject is low and the background is considered to be likely. On the other hand, in the region where the number of blocks in contact with the screen edge is small, the ratio of the number of blocks in contact with the screen edge to the area of the region is small. In this case, the possibility of a background is low, and the first evaluation value is set low because it is considered to be a subject.

  As described above, for example, as shown in FIG. 4C, the regions (1), (7), (8), (9) where the number of blocks in contact with the screen edge is large and the first evaluation value is high. ) As a background and can be removed from the subject area candidates. Further, in the above method, the condition to be regarded as the background is not distinguished between the upper, lower, left and right edges of the screen, but the condition may be different at the upper, lower, left and right edges. For example, if it touches the bottom edge of the screen but does not touch the top, left, or right edge, it will not be regarded as the background, or if it touches two or more sides of the top, left, or right edge, it will be regarded as more background. The first evaluation value may be increased.

  In the above method, the block is not in contact with the screen edge, but it is difficult to determine the region far from the center of the screen as the background. Therefore, as another calculation method of the first evaluation value, the calculation may be performed based on whether each region exists near the screen edge. In this case, the arrangement of the blocks constituting the area and the barycentric position of the area are calculated, and it is determined from the result whether the area exists near the screen edge. If it exists near the edge of the screen, the first evaluation value is set high. In this way, as shown in FIG. 4D, the areas (2) and (5) that are not in contact with the edge of the screen but are separated from the center of the screen are regarded as the background and are removed from the candidate object areas. Is possible.

  With these methods, the first evaluation value can be calculated for each region, and the region with the high first evaluation value can be regarded as the background region. Furthermore, an area having the same color information and luminance information as the area removed as the background is regarded as the background. For example, the average value of the luminance, hue, and saturation of each region is calculated, and the difference between the average values of the background-removed region and the non-removed region is calculated. When this difference is small, even an area that is not removed as a background is regarded as a background and is removed from the subject area candidates.

  Furthermore, the threshold value that the difference is considered to be small may be changed according to the color. For example, it is considered that green such as grass is more likely to be the background than the subject, and the threshold value is increased so that the green region is regarded as the background even if the difference is relatively large. In this way, as shown in FIG. 4E, it is possible to consider the background-like areas (3) and (6) as the background even if they exist near the center of the screen.

  Next, in step S303, an area to be regarded as a subject is determined from the areas not removed as the background in step S302. As an example, a method will be described in which an evaluation value (hereinafter referred to as a second evaluation value) representing the likelihood of a subject in each region is calculated and a region regarded as a subject is determined according to the value.

  As the second evaluation value, the following weights Weight_1 to Weight_4 are used. The weight Weight_1 is a weight according to the position of the center of gravity of the region in the image, and the weight Weight_2 is a weight according to the size of the region. The weight Weight_3 is a weight according to whether the shape of the region is close to an aspect ratio of 1: 1, and the weight Weight_4 is a weight according to whether the shape of the region is close to a circle.

  For example, the closer the center of gravity of the region is to the center of the screen, the more likely it is to be a subject, and the value of the weight Weight_1 is increased. Also, the larger the area, the more likely the subject is, and the value of the weight Weight_2 is increased. On the other hand, whether or not the shape of the region is close to an aspect ratio of 1: 1 is determined as follows. First, the distance from the centroid position of the area to the farthest point of the area is calculated, and a circumscribed circle is drawn with the line segment connecting the centroid position and the farthest point as the center and the radius as the center. Then, as the ratio of the area of the area to the area of the circumscribed circle is larger, the aspect ratio of the subject is considered to be closer to 1: 1, and the value of the weight Weight_3 is increased. Whether or not the shape of the region is close to a circle is determined by calculating the length of the outer periphery of the region and calculating the ratio of the area of the region to the length of the outer periphery. If the area ratio is large, it is considered to be close to a circle, and the value of the weight Weight_4 is increased.

  As described above, these weights Weight_1 to Weight_4 are calculated for each region not removed as a background. Then, the product of the weights Weight_1 to Weight_4 is used as the second evaluation value, and the area having the largest second evaluation value among the areas is determined as the subject area. In the example shown in FIG. 4E, the regions (4), (10), and (11) are regions that have not been removed from the candidates as backgrounds. In this case, it can be said that the area (4) is closer to the center of the screen than the areas (10) and (11), the size of the area is large, the shape is close to 1: 1, and the shape is close to a circle. . For this reason, the evaluation value of the subjectivity of the area (4) is the highest, and it is determined as the subject area.

  Here, as a condition for determining the subject area, a condition that the second evaluation value is equal to or higher than a preset threshold value may be further added to the condition that the second evaluation value is the highest. Further, the second evaluation values of the plurality of regions are compared, and when there is a large difference between the evaluation value of the region where the second evaluation value is the largest and the evaluation value of the other region, There is a high possibility that the subject is prominent and prominent. In such a case, the region having the largest second evaluation value is easily regarded as the subject region.

  In the above-described method, the subject region is determined from one image, but the subject region may be determined using a plurality of images that are continuous in time series. For example, the subject area is set only when the same area continuously has the second largest evaluation value. Further, every time the same area is continuously detected, a weight to be increased is added to the second evaluation value, and when it is equal to or more than a set threshold value, it is determined as a subject area. By the above method, it is possible to detect the subject area from the image.

<Execution of subject tracking>
Next, a method for tracking the subject in step S203 in FIG. 2 will be described. FIG. 5 is a flowchart illustrating an example of a processing procedure for tracking a subject by the subject tracking unit 105.
First, in step S501, it is determined whether it is time to start tracking a subject. If it is determined that it is time to start tracking the subject, the process proceeds to step S502. If tracking of the subject has already started, the process proceeds to step S503.

  In step S502, a subject tracking region is set based on the attention point calculated by the subject region attention point calculation unit 103. As an example, a case where an image is divided into 16 × 12 blocks will be described. First, a characteristic color is searched based on color information, centering on a block where a point of interest exists. The condition for the feature color is a condition that there is a block of colors with a certain threshold area, and there are few blocks of the same color in the image. That is, color information that exists only in the subject area and does not exist in the background area is used as the characteristic color. By doing so, it is possible to suppress erroneous tracking that tracks an area other than the subject, and it is possible to accurately track only the subject. When a characteristic color is found, the color information of the peripheral blocks of the block having the characteristic color is stored in a memory or the like (not shown) and used for subject tracking. Here, the peripheral blocks are, for example, four blocks adjacent to each other in the vertical and horizontal directions of a block having a characteristic color.

  Next, in step S503, subject tracking is executed based on the subject region to be tracked. Here, a case where an image is divided into a plurality of blocks as in step S502 will be described. First, a block that matches the characteristic color calculated in step S502 is searched for in a specific area of the image. Whether the color matches the feature color is calculated by, for example, calculating the difference between the feature color and each of R, G, and B of the target block.

  If there is a block that matches the feature color, it is further determined whether or not the neighboring blocks of the target block also match from the differences of R, G, and B. At this time, the evaluation value is calculated so as to increase as the difference between R, G, and B of the target block decreases and as the number of matching blocks among the peripheral blocks increases. After searching for a block whose evaluation value is greater than or equal to the threshold value within a specific area in the image, calculate the distance between the subject area detected in the previous tracking and each block having an evaluation value equal to or greater than the threshold value. The evaluation value is multiplied by a weight that increases. A block having the highest evaluation value (hereinafter referred to as a third evaluation value) multiplied by this weight is regarded as a tracking destination area of the subject area.

  Although it is possible to perform subject tracking by the above-described method, the color of the subject area set when tracking is started during subject tracking in order to accurately track the subject following changes in the subject. Information such as distribution may be updated. Also, the image is divided into a plurality of blocks and subject tracking is executed based on the information. After determining the block having the highest third evaluation value, the tracking start area in the higher-resolution image is determined. Differences such as color information may be further compared. By doing so, it becomes possible to make the subject tracking performance more accurate.

  Next, in step S504, it is determined whether subject tracking is to be performed again for the tracking destination area determined in step S503. For example, if it is determined that there is a high possibility that the tracking destination area calculated by subject tracking is not a subject, subject tracking is not continued.

  As an example of a method for determining whether to perform subject tracking, an evaluation value (hereinafter referred to as a fourth evaluation value) is calculated from a plurality of items, and the determination is made according to the value. When calculating the fourth evaluation value, first, R, G, and B differences between the subject area at the start of tracking and the subject area being tracked, the size of the subject area, and the distance from the screen center of the subject area , And the weights for the four items of time from the start of subject tracking. Then, a fourth evaluation value is calculated by adding these four items.

  For example, the difference between R, G, and B of each block is compared between the subject area at the start of tracking and the subject area being tracked as color information, and the larger the difference, the larger the variation of the subject, and the non-subject area Since there is a possibility of tracking, the weight is reduced. As for the size of the subject region in the image, the maximum size from the start of subject tracking is compared with the current size. Then, as the current size becomes smaller than the maximum size, the subject area may not be the main subject in the image as compared to when tracking is started, so the weight is reduced.

  In addition, as the distance from the center of the subject area to the screen center increases, there is a possibility that the subject is not the main subject as well, so the weight is reduced. Further, with respect to the time after the start of subject tracking, the longer the time, the more likely it is that the subject cannot be tracked with high accuracy, so the weight is reduced.

  As described above, by calculating the fourth evaluation value, it is determined whether or not to perform subject tracking again. When the fourth evaluation value is large, the subject tracking accuracy is high, and the tracking is continued to be regarded as the main subject. On the other hand, when the fourth evaluation value is small, the subject tracking accuracy is low and there is a possibility that the subject is no longer the main subject, so the tracking is stopped. By the above method, it is possible to track the subject.

<Correction of attention point position>
Next, a method of calculating and correcting the attention point of the subject area detected by subject detection in step S209 in FIG. 2 will be described. FIG. 6 is a flowchart illustrating an example of a processing procedure for calculating and correcting the attention point.
First, in step S601, the subject area attention point calculation unit 103 calculates the attention point from the blocks constituting the subject area. As a point of interest representative of the position of the subject area, for example, a block including the position of the center of gravity of the subject area is used.

  In step S602, the subject area attention point correction unit 104 determines whether the subject area includes a block to which the attention point belongs. For example, if an elliptical gray area shown in FIG. 7A is a subject area, the central black dot is the center of gravity position. On the other hand, as shown in FIG. 7B, when the vicinity of the center of the subject area is hollow, the position of the center of gravity exists in the hollow area.

  Here, in order to determine whether a block including the center of gravity position is included in the subject area, color information and luminance information of the block including the center of gravity position and the subject area are compared. If the difference in luminance value, saturation value, and hue value between the subject area and the block including the center of gravity is within the threshold value, it is determined that the block including the center of gravity is included in the subject area, and the position of the target point Is not corrected. Therefore, in this case, the process is terminated as it is. On the other hand, if the threshold value is exceeded, it is determined that the block including the barycentric position is not included in the subject area, and correction is required so that the position of the target point is located within the subject area. Therefore, in this case, the process proceeds to step S603.

  In step S603, the subject region attention point correction unit 104 starts a process of acquiring block information on the upper, lower, left, and right of the attention point. Hereinafter, the processing in steps S603 to S605 is repeated until all the upper, lower, left, and right block information is acquired. In this way, the correction information is determined by sequentially comparing the upper, lower, left and right block information.

  In step S604, the subject region attention point correction unit 104 calculates a width OutsideLength [n] that is not the subject region in the four directions of the attention point. In the example shown in FIG. 7 (c), as indicated by the thin line arrows, there are one block that is not the subject area above and to the left of the point of interest, and there is no block in the right and down directions. That is, OutsideLength [1] = OutsideLength [3] = 1, OutsideLength [2] = utsideLength [4] = 0. Here, [n] is 1 on the top, 2 on the bottom, 3 on the left, and 4 on the right.

  Next, in step S605, the width InsideLength [n] of the subject region outside the region that is not the subject region is calculated in four directions, up, down, left, and right of the target point. In the example shown in FIG. 7C, there are two blocks above and below the point of interest, and three blocks left and right as indicated by thick arrows. That is, InsideLength [1] = InsideLength [2] = 2, InsideLength [3] = InsideLength [4] = 3.

  When the processing in steps S603 to S605 is repeated to acquire the vertical and horizontal block information, in step S606, it is determined in which direction the attention point is shifted in the four directions of the attention point. For example, the method for determining the direction of shifting is as follows.

  First, only the direction in which the value of the width InsideLength [n] is greater than or equal to the threshold value is selected. When the inside of a narrow subject area is set as a point of interest, when tracking is performed using information around the point of interest, most of the tracked region is occupied by a background portion other than the subject region. Then, there is a possibility of tracking using the information of the background portion, and erroneous tracking that tracks to a place other than the subject area is likely to occur. Therefore, a position where there is a wide subject area in the vicinity is set as a point of interest.

  Next, among the directions in which the value of the selected width InsideLength [n] is equal to or greater than the threshold, the direction having the smallest value of the width OutsideLength [n] is selected. This is to prevent the position detected by subject detection and the position tracked by subject tracking from shifting as much as possible while suppressing erroneous tracking. If there are two or more directions in which the width OutsideLength [n] is the smallest, the direction in which the width InsideLength [n] is large is selected. Further, when there is only a direction in which the value of the width InsideLength [n] is equal to or less than the threshold value, the shape of the subject area may be complicated. In that case, since subject tracking is difficult to execute, subject detection is performed again. In the example shown in FIG. 7C, the right direction in which the value of the width InsideLength [n] is large and the value of the width OutsideLength [n] is small is the direction in which the point of interest is shifted.

  Next, in step S607, correction for shifting the attention point in the direction determined in step S606 is performed. In the case of the example shown in FIG. 7D, the attention point is corrected in the right direction. At this time, the correction position of the target point is the central block of the subject area existing in the correction direction.

  With the above method, according to the present embodiment, when the position of the point of interest is outside the range of the detected subject area, the point of interest is corrected, so even if the subject has a donut shape, it is accurate. The subject can be tracked.

  Furthermore, since the subject information display unit 106 displays information indicating the subject such as a rectangular frame centered on the subject region in the image, the region being tracked for the user to be used can be visually represented. Is possible. Here, when the correction amount of the attention point is large, the actual subject area and the display position of the subject information are greatly shifted. Therefore, tracking of the subject is performed at the corrected attention point, and the correction amount depends on the correction amount. Displays information indicating the subject using the point of interest before correction.

(Other embodiments)
The present invention can also be realized by executing the following processing. That is, software (program) that realizes the functions of the above-described embodiments is supplied to a system or apparatus via a network or various storage media, and a computer (or CPU, MPU, or the like) of the system or apparatus reads the program. It is a process to be executed.

102 Subject region specifying unit 103 Subject region attention point calculation unit 104 Subject region attention point correction unit 105 Subject tracking unit

Claims (9)

Identifying means for identifying the subject area from the input image;
Calculating means for calculating a point of interest representative of the position of the subject area specified by the specifying means;
Tracking means for tracking the subject area based on the attention point calculated by the calculation means;
When the region including the attention point is a region different from the subject region, the calculation unit corrects the position of the attention point within the subject region,
The image recognition apparatus characterized in that the tracking means tracks the subject area based on a point of interest at the corrected position.   The image according to claim 1, wherein the specifying unit specifies a subject region using at least luminance information or color information in the image, or specifies a subject region in accordance with an operation by a user. Recognition device.   The image recognition apparatus according to claim 1, wherein the attention point is a gravity center position of the subject area.   The image recognition apparatus according to claim 1, wherein the tracking unit tracks the subject area using at least luminance information or color information in the image.   5. The calculation unit according to claim 1, wherein the calculation unit determines whether the regions are different from each other based on a difference in luminance information or color information between at least the region including the attention point and the subject region. The image recognition device according to any one of the above.   The identification means calculates an evaluation value representing the likelihood of a subject from the position, size, and shape of the region in the image, and identifies the subject region based on the calculated evaluation value. The image recognition device according to any one of 1 to 5. Display control means for displaying on the display means together with the image that the subject exists in the subject area tracked by the tracking means;
The display control means further displays on the display means that the subject exists using the attention point before or after the correction by the calculation means. Item 7. The image recognition device according to any one of Items 1 to 6. A specifying step of specifying a subject area from the input image;
A calculation step of calculating a point of interest representative of the position of the subject area specified in the specifying step;
A tracking step of tracking the subject area based on the attention point calculated in the calculation step,
In the calculation step, when the region including the attention point is a region different from the subject region, the position of the attention point is corrected inside the subject region,
In the tracking step, the subject region is tracked based on the attention point of the corrected position. A specifying step of specifying a subject area from the input image;
A calculation step of calculating a point of interest representative of the position of the subject area specified in the specifying step;
Causing the computer to execute a tracking step of tracking the subject area based on the attention point calculated in the calculation step;
In the calculation step, when the region including the attention point is a region different from the subject region, the position of the attention point is corrected inside the subject region,
In the tracking step, the subject area is tracked based on a point of interest at the corrected position.

Images