JP2014142832A - Image processing apparatus, control method of image processing apparatus, and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve recognition accuracy of a subject.SOLUTION: An image processing apparatus includes an acquisition unit that acquires distance information to an imaging object, an image input unit that acquires an image of the imaging object as an input image, an image creation unit that creates one or more conversion images obtained by magnifying the size of each of areas of the input image on the basis of the distance information, and a detection unit that detects a specific object from the input image by using the conversion images.

Description

  The present invention relates to an image processing device, a control method for the image processing device, and a program.

  Conventionally, many methods for detecting a specific object using a pyramid image obtained by enlarging or reducing an input image have been proposed.

  In Patent Document 1, the size value of a target object candidate area in an image input via an image input unit that inputs an image obtained by photographing a predetermined area to an image processing apparatus is detected. Then, a magnification for enlarging / reducing the image is determined by the size value of the detected object candidate area and a reference value set in the discriminator, and the image is enlarged / reduced by the magnification. An image within the frame is extracted by moving the image corresponding to the reference value on the enlarged / reduced image, and an image within the frame and a face detection classifier capable of detecting a face corresponding to the reference value. It has been proposed to perform face detection.

  In Patent Document 2, the pyramid image generation unit reduces or enlarges the captured image by a magnification set in advance according to the distance from the camera that performs imaging to the subject to be detected, and uses the pyramid to detect the subject. Generate an image. There has been proposed a method in which the detection region determination unit determines a detection region for detecting a subject from all regions on the pyramid image, and the subject detection unit detects a subject from the detection region.

JP 2008-191760 A JP 2011-53915 A

  However, the configuration of Patent Document 1 has a problem that a luminance difference does not occur in the case of a telephoto captured image using a high pixel sensor because illumination does not reach. In addition, when the background reflects too much light, such as rain or snow, it is difficult to measure the brightness difference.

  Further, the configuration of Patent Document 2 has a problem that a pyramid image of a captured image cannot be appropriately selected unless the angle of view between the entire captured image of the natural image and the entire distance map (also referred to as a depth map) is the same. Further, in the case of a telephoto captured image using a high pixel sensor as in Patent Document 1, there is a problem that the parallax of the range finder becomes large.

  SUMMARY OF THE INVENTION In view of the above problems, an object of the present invention is to provide an image processing apparatus capable of selecting a pyramid image more appropriately and improving subject recognition accuracy through cooperation between the image processing apparatus and a distance sensor. .

An image processing apparatus according to the present invention that achieves the above object is as follows.
An acquisition means for acquiring distance information with respect to an imaging target;
Image input means for acquiring the image to be photographed as an input image;
Image generating means for generating one or more converted images obtained by scaling the size of each area of the input image based on the distance information;
Detecting means for detecting a specific object from the input image using the converted image;
It is characterized by providing.

  According to the present invention, it is possible to reduce erroneous detection of a subject and to perform faster subject detection.

1 is a block diagram showing a configuration of an image processing apparatus according to a first embodiment. The figure which performs a human body detection in the detection object reduction / enlargement conversion image which expanded and reduced the input image of the image processing apparatus which concerns on 1st Embodiment. The figure of the input all pixel read data of the image processing device concerning a 1st embodiment. The depth figure as a distance map from the distance sensor which concerns on 1st Embodiment. The figure of the table | surface which took the correspondence of the resolution etc. of the all-pixel read-out image and depth map which concern on 1st Embodiment, and analyzed the search layer in each area | region. 3 is a flowchart showing an operation procedure of the image processing apparatus according to the first embodiment. The block diagram which shows the structure of the image processing apparatus which concerns on 2nd Embodiment. The figure of the input all pixel read data at the time of using a wide angle optical system with the image processing apparatus which concerns on 2nd Embodiment. The figure which shows each layer image produced from the image which extracted the human body candidate area | region after moving body area | region detection with the image processing apparatus which concerns on 2nd Embodiment. 9 is a flowchart showing an operation procedure of the image processing apparatus according to the second embodiment. The figure of the positional relationship of the all-pixel read-out image and depth diagram which concerns on 2nd Embodiment.

(First embodiment)
The first embodiment of the present invention will be described below with reference to FIGS. FIG. 1 is a block diagram of an image processing apparatus 10 according to the first embodiment. The image processing apparatus 10 includes an image input unit 11, a storage unit 12, an input unit 13, a distance sensor 14, a distance sensor input unit 15, a human body detection unit 16, an image size scaling unit 17, and an external output. And the operation is controlled by a CPU (not shown).

  The image input unit 11 has a function of cutting out only a part and acquiring it as an input image when acquiring an image from an image sensor or the like. The storage unit 12 stores setting values of sensors and the like connected to the image input unit 11, stores area information instructed to the image input unit 11 and the distance sensor input unit 15, and stores various parameters. . The input unit 13 inputs various setting values and area designation. The distance sensor 14 has a distance measuring function, and measures the distance from the image processing apparatus 10 to the photographing target. The distance sensor input unit 15 acquires data from the distance sensor 14 to create a depth map, and performs alignment correction so as to match the processing of the image processing apparatus 10.

  When detecting a human body as a specific object, the human body detection unit 16 detects a human body in a plurality of images obtained by reducing and enlarging an input image using a reference human body matching pattern. The image size scaling unit 17 generates an image obtained by converting the size of the input image to a designated arbitrary size or a designated arbitrary magnification. The external output unit 18 outputs the result detected by the human body detection unit 16 to the outside.

  Here, the distance sensor 14 may be various distance sensors according to the distance measuring method. In the present embodiment, a distance sensor that irradiates an infrared dot pattern, captures and analyzes the situation with an infrared camera, and extracts distance information is used. Further, it is assumed that the image processing apparatus 10 according to the present embodiment is in a position that is substantially integrated with an image input apparatus (not shown) connected to the image input unit 11. Therefore, the distance from the image input device (not shown) can be regarded as the same as the distance measured by the distance sensor 14.

  In this embodiment, it is assumed that a human body is detected as a specific object. Therefore, the human body detection unit 16 performs the following process to detect human bodies having different sizes.

  First, a plurality of images (called layer images) obtained by reducing or enlarging the input image at a preset magnification so that the size of the human body candidate region matches the reference human body matching pattern is an image size scaling unit. 17 is generated. Then, the human body is detected by performing pattern matching between the layer image and the matching pattern.

  With reference to FIG. 2, an outline of the human body detection process performed by the image processing apparatus 10 according to the first embodiment will be described. In FIG. 2, reference numeral 21 denotes a reference human body collation pattern. A human body is detected by performing line scanning on the target layer image for the matching pattern 21. In the present embodiment, a plurality of images obtained by reducing or enlarging an input image clipped by specifying an area from the image input unit 11 by the image size scaling unit 17 at a preset magnification is used. The reduced, enlarged images 20, 22, and 23 are called layer images. A layer image refers to an image obtained by repeatedly reducing (or enlarging) an input image at a preset magnification. Here, the image 20 is called layer 1, the image 22 is called layer 2, and the image 23 is called layer 3.

  As shown in FIG. 2, it is possible to detect human bodies of different sizes by performing line scanning on the reference human body matching pattern 21 for each layer image. In general, an image (layer image) reduced or enlarged at a preset magnification is generated until the size of the reference human body matching pattern 21 and the size of the input image are substantially the same. Then, the human body detection process is repeatedly performed using the plurality of layer images while replacing the layer images.

  In the present embodiment, the process of reducing and converting the input image until the size of the matching pattern 21 of the human body as a reference and the size of the input image is substantially the same with a preset magnification is performed. Instead, only the reduced image with the selected magnification is generated by the image size scaling unit 17. Details will be described later.

  FIG. 3 is a diagram illustrating an example of all-pixel read data in the image input unit 11 according to the first embodiment. The gray part is an area that schematically represents a road. It is a figure at the time of imaging with the image input device (not shown) toward the diagonal right direction from the diagonal left back. As shown in the image 30, by dividing the image data read out of all pixels into, for example, three vertical and four horizontal areas, it is possible to reduce the processing load that the human body detection unit 16 performs at one time. Number 31 indicates a number assigned to each divided image area for convenience. Human body candidates 32 and 33 indicate human body candidates having different sizes.

  Here, it is assumed that areas where human bodies are likely to appear and human bodies are to be detected are, for example, four areas 2.2, 2.3, 3.2, and 3.3, set in the storage unit 12 through the input unit 13, and stored. Keep it. By doing so, the processing area of the human body detection process in the human body detection unit 16 can be limited. Therefore, it is possible to perform human body detection processing at higher speed and with reduced false detection.

  FIG. 4 is an example of a depth diagram obtained by mapping data measured by the distance sensor according to the first embodiment. The image 40 is a depth diagram input from the distance sensor 14. The color intensity represents the closeness of the distance. That is, a is the closest and f is the farthest. The number 41 indicates the number assigned to the area when the depth map is divided into 3 × 4 in the vertical direction following the image 30 which is the all-pixel read data. When the angle of view between the image 40 and the all-pixel read data is almost the same as in the image 40, the same vertically and horizontally divided region can be referred to as the depth information of the image region of the image 40. In this example, the depth information 42 is depth information corresponding to the human body candidate 32, and the depth information 43 is depth information corresponding to the human body candidate 33. In this example, in FIGS. 3 and 4, Arabic numerals 1 to 4, which are numbers assigned to the respective areas, correspond to Roman numerals I to IV, respectively.

  However, there are cases where the positional relationship between the all-pixel read data and the depth map is different, or the distance accuracy that can be detected is not the same as the area imaged as the all-pixel read data. This case will be described in detail in the second embodiment.

  FIG. 5 shows the correspondence between the position and resolution of the image 30 of the all-pixel read data and the image 40 of the depth map according to the first embodiment, and analyzes the search layer image of a specific object (human body in this embodiment). It is the table | surface which put together the computed result.

  Human body detection processing corresponding to depth information (Depth) for each region (Region) of the image 30 of all pixel readout data based on the depth information of the image 40 of the depth diagram corresponding to each region of the image 30 of all pixel readout data The result of calculating the layer image (Layer) is shown in the table. The human body detection processing layer image uses six layer images 1 to 6 that are numbered 1 in the larger image and reduced by five levels with a preset magnification. The relationship shown in this table may be set as a fixed value, but it may be calculated in real time and the value may change every moment. Details will be described later with reference to the flowchart of FIG.

  Here, FIG. 6 is a flowchart showing an operation procedure of the image processing apparatus 10 according to the first embodiment.

  In FIG. 6, first, a maximum input field angle (resolution) as a parameter of an image input device (not shown) connected to the image processing device 10 from the input unit 13 and a partial field angle region as a field angle of interest. The data is input and stored in the storage unit 12. Here, the partial angle of view area is set as an image partial cutout area in, for example, a CMOS sensor as an image input device. With the recent increase in the number of pixels in an image sensor, the load of image processing due to readout of all pixels (maximum input angle of view readout) has increased. By using partially cut out images, the load of image processing has been reduced. The processing speed can be improved. In addition, various distance measurement parameters related to the distance sensor 14 such as the vertical and horizontal sizes of the depth map of the distance sensor 14 and the longest search distance necessary for calculation of depth information in the distance sensor input unit 15 are stored in the storage unit 12 by the input unit 13. Set and save (S100).

  The distance sensor input unit 15 creates a depth map image 40 that is a distance map from various distance measurement data input from the distance sensor 14 and stores it in the storage unit 12 (S101).

  Next, the following processing is executed based on various distance measurement parameters of the distance sensor 14 set by the input unit 13 and the maximum input field angle of the image input device (not shown). That is, it is determined whether or not the image 40 of the depth map as the distance map that is distance information corresponds to the image 30 of the all-pixel read data as the input image (S102).

  If it is determined that they are compatible (S102; YES), the alignment of the image 30 of all-pixel read data and the image 40 of the depth map is performed (S103). On the other hand, if it is determined that they are not compatible (S102; NO), the process proceeds to S105.

  After performing the alignment, the human body detection unit 16 includes the depth map image 40 as a distance map stored in the storage unit 12, and the maximum input field angle (resolution) information from the image input device (not shown). Based on the above, the following processing is performed. That is, the human body detection processing layer (Layer) corresponding to the depth (Depth) is generated for each region (Region) of the image 30 of the all-pixel read data in FIG. 5, and the layer correspondence table for each generated region is stored in the storage unit 12. (S104).

  The creation of the correspondence table is calculated by using the feature that the human body as the specific object is imaged as the distance from the image input device (not shown) to the object to be photographed increases (depth increases). . Further, the calculation is performed using a feature that a human body is imaged closer to the ground around the image input device (not shown). For each input image region, a smaller layer image (converted image) is generated as the distance information is closer, and a larger layer image (converted image) is generated as the distance information is farther away.

  In the present embodiment, the variation in the size of the detected human body is taken into consideration in the region where the input image is equally divided, and a variation width is given to the layer image selection. For example, R (1, 1) in FIG. 5 has a depth Depth as distance information of f, which is quite deep, that is, the distance is far. A human body as a specific object that is imaged at a far distance is characterized by being relatively small. The human body candidate region captured in a small size is more easily detected by using a layer image that does not reduce (or enlarge) the image in the human body detection process performed by pattern matching the reference human body matching pattern 21 in FIG. . Therefore, a large layer image 1-2 is selected for R (1, 1) as a region to be cut out from the input image, and an image is generated. In this way, based on the distance information acquired from the distance sensor 14, one or more converted images obtained by scaling the size of the area are generated for each area of the input image.

  In the present embodiment, an area obtained by equally dividing the input image is used. However, a specific area at an arbitrary position is specified, and a human body detection layer (Layer) image corresponding to the depth (Depth) is selected to select an image. You may make it produce | generate.

  Further, the image input unit 11 reads out the target partial angle of view area information stored in the storage unit 12 by the input unit 13 and acquires the information from the image input device (not shown) using the target partial angle of view area information. The partial image of interest is cut out from the obtained image and acquired (S105).

  The image input unit 11 obtained by cutting out and acquiring the target partial image sends the target partial image to the human body detection unit 16. The human body detection unit 16 that has received the target partial image executes the following processing. That is, the human body detection target layer image is selected from the layer image correspondence table as shown in FIG. 5 (S106). If it is determined No in S102, the layer correspondence table has not been created, and therefore all layer images are selected.

  Based on the layer image selected in S106, a converted image is generated by the image size scaling unit 17 as described with reference to FIG. 2 from the target partial image acquired in S105, and each selected layer image is generated. To do. Thereafter, human body detection is performed as described in FIG. 2 (S107).

  Then, it is determined whether or not a human body is detected by the human body detection unit 16 (S108).

  When a human body is detected by the human body detection unit 16, the detection result is notified to the outside via the external output unit 18 (S109). Thereafter, it is determined whether or not to end the human body detection scan (S110). If not, the process returns to S101 and repeats the process.

  As described above, the processing load can be reduced by appropriately selecting a reduced image using the distance information acquired from the distance sensor.

(Second Embodiment)
FIG. 7 is a block diagram of an image processing apparatus according to the second embodiment. The image processing apparatus 70 includes an image input unit 11, a storage unit 12, an input unit 13, a distance sensor input unit 15, a human body detection unit 16, an image size scaling unit 17, an external output unit 18, and a moving object. An area detection unit 19 and a geometric correction unit 20 are provided. In the present embodiment, the distance sensor 14 is configured separately from the image processing apparatus 70. The image processing device 70 further includes a moving object region detection unit 19 and a geometric correction unit 20 in addition to the components other than the distance sensor 14 in the configuration of the image processing device 10 according to the first embodiment.

  The moving object region detection unit 19 detects a moving object in the input image. The geometric correction unit 20 performs geometric correction of the input image using a geometric conversion parameter based on the optical characteristics. Since other components are the same as those in the first embodiment, description thereof is omitted.

  FIG. 8 is a diagram showing input image data from an image input apparatus (not shown) including the wide-angle optical system according to the second embodiment. This input image data is a diagram taken by directing the image input device in the same direction as in FIG. 3 in the first embodiment, and the gray distorted arc portion is a region schematically representing a road.

  In the image 80 of all-pixel read data, the moving object region information 81 indicating a moving object detection frame when a single moving object is detected by the moving object region detection unit 19, and the moving object is a specific object (here, human body) candidate A human body candidate region 82 indicating that the In this embodiment, since the input image is assumed to be an image input device (not shown) using a wide-angle lens, the peripheral portion of the image is distorted and imaged. The image distortion can be converted and generated by correcting the distortion by correcting the distortion by geometric correction by the geometric correction unit 20 using the geometric conversion parameter based on the optical characteristics. In addition, since an image input device (not shown) using a wide-angle lens may be disposed away from the image processing device 70, the distance sensor 14 can measure the distance from the vicinity of the image input device (not shown). Is configured to be out of the image processing apparatus 70.

  FIG. 9 shows each layer image created from an image obtained by cutting out a certain human body candidate area after moving object area detection in the image processing apparatus according to the second embodiment. FIG. 9 shows the human body candidate area 82 in FIG. 8 cut out by the image cutout acquisition function of the image input unit 11 and the cut out partial input image subjected to geometric correction by the geometric correction unit 20 using the geometric transformation parameter based on the optical characteristics. is there. Then, the geometrically corrected partial input image (human body candidate region image 90) is reduced or enlarged by the image size scaling unit 17 at a magnification set in advance through the input unit 13, and a plurality of images (a plurality of layer images) are obtained. ). In the example of FIG. 9, four layer images 9-3 to 9-6 are created.

  Here, FIG. 10 is a flowchart showing an operation procedure of the image processing apparatus according to the second embodiment. Hereinafter, with reference to FIG. 4 for a depth map as a distance map, the operation procedure of the image processing apparatus 70 will be described in detail with reference to FIGS.

  First, the maximum input angle of view (resolution), the angle of view after geometric correction, and the geometric correction of an image input device (not shown) provided with a wide-angle optical system connected to the image processing device 70 from the input unit 13. The geometric transformation parameters based on various optical characteristics for the above are input and stored in the storage unit 12 (S200). The process of S201 is the same as the process of S101.

  Next, the following processing is performed based on various distance measurement parameters of the distance sensor 14 set by the input unit 13, the maximum input angle of view of the image input device (not shown), and the angle of view after geometric correction. Run. That is, the distance sensor input unit 15 is a geometrically corrected image (not shown) obtained by geometrically correcting the depth map image 40 as a distance map, which is distance information, and the image 80 of all-pixel read data as an input image by the geometric correcting unit 20. Is determined) (S202).

  When it is determined that the image can be handled, the geometric correction image (not shown) obtained by geometric correction of the image 80 of the all-pixel read data by the geometric correction unit 20 is aligned with the image 40 of the depth map (S203). On the other hand, when it is determined that correspondence is not possible, positioning is not performed and the depth map is indefinite.

  Further, the image input unit 11 reads the moving body region information 81 detected by the moving body region detection unit 19, regards the moving body region information 81 as a human body candidate region, and executes the following process. That is, a human body candidate area image 82 (human body candidate area partial image) is cut out and acquired from an image input device (not shown) (S204).

  Here, as shown in FIG. 8, the human body candidate area image 82 is larger in size than the moving body area information 81 because it must be larger than the margin of the reference human body collation pattern in human body detection. In addition, since the image size may change during the next geometric correction, it is necessary to cut out the human body candidate area image 82 larger than the moving body area information 81 by looking at the margin.

  Thereafter, the geometric correction unit 20 performs geometric correction on the human body candidate region partial image using the geometric conversion parameters based on various optical characteristics for geometric conversion set by the input unit 13, and generates geometric corrected image data. To do. Then, both the geometrically corrected image data and the moving object position information are transmitted to the human body detection unit 16 (S205).

  Here, consider a case where the human body candidate area image 90 and the moving body position information that have been geometrically corrected are received as shown in FIG. The human body detection unit 16 performs the following processing. That is, the depth map image 40 as a distance map is read from the storage unit 12, and it is determined whether or not the moving object position information of the human body candidate region image 90 is within the range of the depth map image 40 (S206). .

  When the moving body position information of the human body candidate region image 90 is within the range of the depth map image 40, the depth map image 40 aligned by the distance sensor input unit 15 is read from the storage unit 12, and this depth map image is displayed. 40, the human body detection target layer image is selected based on the human body candidate region image 90 and the moving body position information. The human body detection target layer image is selected using the feature that the human body is captured with a smaller size as the distance from the image processing apparatus increases, and the details will be described later. Thereafter, based on the selected human body detection target layer image, the converted image is generated from the human body candidate region image 90 by the image size scaling unit 17 as described with reference to FIG. Thereby, each selected layer image is generated (S207).

  In the present embodiment, each selected layer image is converted and generated by the image size scaling unit 17 as four layer images 9-3 to 9-6 as shown in FIG. The four layer images are selected and generated for the following reason. From the information of the human body candidate area image 90 that has been imaged and geometrically corrected and the image 40 of the depth map, it can be seen that the human body candidate area image 90 exists at a short distance. A human body as a specific object present at a short distance is characterized by being captured relatively large. Therefore, when performing human body detection by performing line scanning with the reference human body matching pattern 21, an image obtained by reducing the human body candidate area image 90 to a smaller size is more likely to match the reference human body matching pattern 21. Therefore, four layer images 9-3 to 9-6 obtained by reducing the human body candidate region image 90 to be relatively small are converted and generated by the image size scaling unit 17. Then, using the four layer images, the human body detection process is performed as described in FIG.

  On the other hand, the case where the moving body position information of the human body candidate region image 90 is outside the range of the image 40 of the depth diagram is a case as shown in FIG. FIG. 11 is a diagram schematically illustrating a case where the geometrically corrected image obtained by geometrically correcting the image 30 of the all-pixel read data of the image input unit 11 by the geometric correcting unit 20 is wider than the image 40 of the depth diagram. . The same applies to the case where the image 30 of the all-pixel read data of the image input unit 11 is wider than the image 40 of the depth map as the distance map as in the first embodiment. The area 110 corresponds to a geometrically corrected image obtained by geometrically correcting the image 30 by the geometric correcting unit 20, and the area 112 corresponds to a depth map image 40 as a distance map.

  As described above, when the moving body position information of the human body candidate area image 90 is outside the range of the image 40 of the depth map, the following processing is performed. That is, all layer images are selected based on the human body candidate area image 90 and the moving body position information, and a converted image is generated from the human body candidate area image 90 by the image size scaling unit 17 as described in FIG. An image is generated (S208). As described above, a layer image (converted image) is generated at all of a plurality of predetermined magnifications for an area of the input image where the corresponding distance information does not exist.

  Next, human body detection processing is performed as described with reference to FIG. 2 using each of the generated selected layer images (S209).

  Then, it is determined whether or not a human body is detected by the human body detection unit 16 (S210).

  When a human body is detected by the human body detection unit 16, the detection result is notified to the outside via the external output unit 18 (S211). Thereafter, it is determined whether or not the human body detection scan is to be ended (S212). If not, the process returns to S201 to repeat the process.

  As described above, according to the present embodiment, erroneous detection of a subject can be reduced, and higher-speed subject detection can be performed.

  The present invention is not limited to each of the above embodiments, and the following modifications are possible.

  In the first embodiment and the second embodiment, the human body is used as the specific object. However, the specific object is not limited to this and may be an object such as a face, a head, a car, or an animal. In the first and second embodiments, a configuration using a semiconductor sensor such as a CMOS sensor or a CCD sensor or another image sensor as the image input device (not shown) may be used.

  In the second embodiment, when a plurality of moving objects are detected, the gaze moving object may be specified by the user, or may be automatically specified under a predetermined condition. The gaze moving object may be switched to another moving object at certain time intervals from the moving object. As the predetermined condition, in order from the inner periphery, in order from the outer periphery, in order of the size of the moving object, or in order of the speed of the moving object, etc. can be considered.

  In the second embodiment, processing is performed using a wide-angle lens input image from an image input device (not shown) provided with a wide-angle optical system. A wide-angle optical system may be used. In the case of an omnidirectional fisheye lens or an omnidirectional mirror, a distance sensor that can measure an omnidirectional distance can be used. In the second embodiment, the moving object region detection unit 19 is configured separately from the image input device (not shown), but the image input device (not shown) also serves as the moving object region detection unit 19. There may be.

(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.

Claims (8)

An acquisition means for acquiring distance information with respect to an imaging target;
Image input means for acquiring the image to be photographed as an input image;
Image generating means for generating one or more converted images obtained by scaling the size of each area of the input image based on the distance information;
Detecting means for detecting a specific object from the input image using the converted image;
An image processing apparatus comprising:   The image generation unit generates a converted image having a smaller size as the distance of the distance information is closer, and generates a converted image having a larger size as the distance of the distance information is farther, for each region of the input image. The image processing apparatus according to claim 1. The distance information is a distance map;
The image processing apparatus according to claim 1, further comprising an alignment unit configured to perform alignment between the distance map and the input image.   The image processing apparatus according to claim 1, further comprising a geometric correction unit that performs geometric correction on the input image based on optical characteristics.   5. The image generation means generates a converted image at all of a plurality of predetermined magnifications for an area of the input image where no corresponding distance information exists. The image processing apparatus according to any one of the above.   The image processing apparatus according to claim 1, wherein the specific object is a human body, a face, a head, a car, or an animal. An image processing apparatus control method comprising an acquisition means, an image input means, an image generation means, and a detection means,
An acquisition step in which the acquisition means acquires distance information with respect to an imaging target;
An image input step in which the image input means acquires the image to be photographed as an input image;
An image generating step in which the image generating unit generates one or more converted images obtained by scaling the size of the region for each region of the input image based on the distance information;
A detecting step in which the detecting means detects a specific object from the input image using the converted image;
A control method for an image processing apparatus, comprising:   The program for making a computer perform each process of the control method of the image processing apparatus of Claim 7.