JP2016053763A - Image processor, image processing method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To solve a problem that a processing time is rapidly increased if an image size becomes large because the entire image becomes an analysis object when characters are attempted to be read from an image picked up by a digital camera, or the like without omission.SOLUTION: An image processor can suppress character area extraction omission to shorten a processing time by including detection means for detecting a feature area from an input image, determination means for determining a first search range in the input image and a second search range different from the first search range on the basis of the position and the size of the detected feature area, reduction means for performing reduction processing of the second search range with a prescribed reduction rate, and extraction means for extracting character areas from an image of the first search range and a reduced image of the second search range subjected to the reduction processing by using data of each pixel constituting the image and the reduced image.SELECTED DRAWING: Figure 6

Description

  The present invention relates to an image processing apparatus, an image processing method, and a program for extracting characters from an image.

  Attempts have been made to facilitate handling of a large amount of images by analyzing the captured images and extracting information. For example, a technique for identifying a captured subject by similar image search or image analysis is being put into practical use. Alternatively, if a monument or signboard with a name such as a landmark is shown in a commemorative photo taken at a sightseeing spot, the imaging location can be specified by character recognition.

  Thus, a technique for extracting character information by detecting a character region from a captured image and recognizing the character is disclosed (for example, Non-Patent Document 1 and Non-Patent Document 2). However, since it is not easy to extract a character region from a captured image having a non-uniform background, Non-Patent Document 1 focuses on the stroke width and specifies the character region. In Non-Patent Document 2, the character pixel block is selected from the pixel block extracted by using MSER (Maximally Stable Extreme Regions) which is one of the pixel block extraction methods. With MSER, pixels that exist stably can be extracted as a pixel block from a large number of binary images generated from a captured image.

  In addition, it is difficult to accurately extract a region of interest from an image that does not limit the imaging conditions because a subject other than the subject to be noticed is reflected or the size of the region to be noticed varies depending on the photograph. Therefore, Patent Literature 1 and Patent Literature 2 are disclosed as methods for limiting the region to be analyzed to a part of the image.

  In Patent Document 1, a face area that is an attention area is detected from an image, a body area is further detected from the position of the face, and an object area in which a landmark appears is extracted from an area excluding the face and the body. . In Patent Document 2, a part of a subject such as a face or a hand is detected, a face, a hand, or a space between a face and a hand is set as a restricted area, and a feature amount is extracted from an area excluding the restricted area of the attention area. did.

  Further, Patent Document 3 is disclosed as a technique for making the size of a region of interest that varies depending on a photograph constant.

  In Patent Document 3, the size of a region estimated as a face is detected from a captured image before being applied to a face detection discriminator, and compared with a predetermined value to calculate an enlargement / reduction ratio.

JP2013-65156 JP2012-53606 JP2008-191760

Epstein et al., “Detecting text in natural scenes with stroke width transform”, CVPR 2010 Conference on Computer Vision and Pattern Recognition, 2963-2970. Merino-Gracia et al., “A head-mounted device for recognizing text in natural senses”, CBDAR 2011 Proceedings of the 4th international concealment.

  However, in the prior art, it is not possible to quickly extract a region to be analyzed from an image without omission. That is, if the detection accuracy of a region of interest (hereinafter also referred to as a feature region) is not sufficient, the region to be analyzed is not included in the feature region or the region derived from the feature region, and there is a problem that an analysis failure occurs. It was.

  In an example of a marathon event, the accuracy of face detection is greatly affected by the participant's hats, sunglasses, disguise, and the like, so that the face area as a feature area may not be detected correctly. In order to extract the bib number to be analyzed without omission, it was necessary to analyze the entire image. As a result, there is a problem that the time required for analysis increases. In particular, in order to increase the analysis accuracy, if the number number area is searched from all over the image, there is a problem that the processing time increases significantly as the number of pixels of the image being processed increases.

  The present invention has been made in view of the above problems, and an object of the present invention is to provide an image processing apparatus, an image processing method, and a program that can suppress omission of extraction of a character region and shorten the processing time of extraction processing. And

  In order to solve this problem, an image processing apparatus according to the present invention has the following configuration. That is, the image processing device includes a detection unit that detects a feature region from the input image, a first search range in the input image, and a first search range based on the detected position and size of the feature region. Determining means for determining a different second search range, reduction means for reducing the second search range at a predetermined reduction ratio, an image of the first search range, and a reduced image of the second search range that has been reduced On the other hand, it is characterized by having extraction means for extracting a character region using data for each pixel constituting each.

  According to the present invention, the extraction of the character area is suppressed by extracting the character area in both the first search range and the second search range, and the character area extraction process is performed by reducing the second search range. As a result, the processing time of the extraction process can be shortened.

Block diagram showing hardware configuration and functional configuration of image processing apparatus An image showing an example of the processing target of the image processing apparatus The flowchart explaining the flow of the process which reads character information from the image in 1st Embodiment. A flowchart for explaining a flow of character region extraction processing according to the first embodiment. The graph which shows the relationship between the size of the image in 1st Embodiment, and the processing time of the extraction process of a character area The flowchart explaining the flow of the detailed extraction process of the character area in 1st Embodiment The figure which shows the result of face detection and search range in 1st Embodiment The figure which shows the image after reduction of the search range in 1st Embodiment. The figure which shows the mask range and extended search range in 1st Embodiment The flowchart explaining the flow of the detailed extraction process of the character area in 2nd Embodiment The figure which shows the example where the marker is printed on the bib

  DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments for carrying out the invention will be described with reference to the drawings.

(First embodiment)
FIG. 1A is an example of a hardware configuration of the image processing apparatus according to the first embodiment. Reference numeral 102 denotes a CPU that performs processing of the apparatus, reference numeral 103 denotes a ROM that stores a control program, reference numeral 104 denotes a RAM that temporarily stores data being processed, and reference numeral 105 denotes an external storage device such as a magnetic disk. The ROM 103 may store a processing program of this apparatus shown in a flowchart described later. A plurality of CPUs 102 may be provided.

  A network interface 106 is connected to a LAN or WAN and communicates with a remote device. Reference numeral 108 denotes an operation unit such as a keyboard for operating the apparatus, and reference numeral 107 denotes a display unit for displaying the state of the apparatus and information for the operator. Like the touch panel liquid crystal monitor, the operation unit 108 and the display unit 107 may be integrated.

  The processing program of this apparatus shown in the flowchart described later is configured to expand what is stored in the external storage device 105 or supplied from the outside via the network interface 106 to the RAM 104 under the control of the CPU 102. It may be.

  These components are arranged on the system bus 101.

  Note that a general-purpose computer may be used as the hardware configuration of the image processing apparatus of the present embodiment.

  FIG. 1B is a block diagram illustrating an example of a functional configuration of the image processing apparatus according to the first embodiment. Assume that an image to be subjected to image processing is captured from an image input unit (not shown) into the image processing apparatus. The face area detection unit 112 detects a face area as a feature area. The search range determination unit 113 determines, as the search range, a range that is likely to include a character region from the detected face region. The search range determination unit 113 can determine a plurality of search ranges according to the image to be processed.

  The image reduction unit 114 performs image reduction processing on the determined search range, and then the image analysis unit 115 performs analysis processing. The image analysis unit 115 includes a character area extraction unit 116 and a character recognition unit 117. The character area extraction unit 116 extracts a character area from the search range. The number of character region extraction units 116 is not limited to one, and a plurality of character region extraction units 116 may be provided. The character recognition unit 117 reads a character from the character area. Character information obtained by analysis by the image analysis unit 115 is associated with the image-processed image and stored in a memory or a storage device.

  FIG. 2 is an example of an image processed by the image processing apparatus of FIG. 1, which is an image of a citizen marathon or road race in which a general athlete participates. In order to provide the captured image in the form of data or a printed photograph to the applicant, it is necessary to be able to search by image number from the image information. By using the image processing apparatus of the present embodiment, it is possible to read the number number of the participant shown in the image illustrated in FIG. 2 and record it in association with the image.

  Next, an outline of processing for reading character information from an image, which is executed by the image processing apparatus according to the present embodiment, will be described with reference to the flowchart of FIG.

  In step S <b> 301, an image to be processed by an image input unit (not shown) is input and stored in the RAM 104. In step S302, the character area extraction unit 116 extracts a character area from the input image. In step S303, the character recognition unit 117 performs character recognition on the image of the character area extracted in step S302. As a result, all character images in the image are converted into character codes and read as character information. In step S304, the character code as the character recognition result is assigned and stored as metadata of the input image. The image to which the metadata is added is stored in the external storage device 105.

  The character region extraction processing in step S302 is processing for analyzing the entire image in units of pixels and detecting a pixel block to extract a character region. The method exemplified in the background art for analyzing the entire image in units of pixels and detecting a pixel block can prevent deterioration in accuracy such as omission of extraction, but tends to increase analysis processing time.

  For example, the pixel block extraction method MSER used in Non-Patent Document 2 is a method in which locations with close luminance values in an image are combined into one region. In this method, places where luminance values are close in the vicinity of each pixel are connected. In order to obtain a place where the luminance value is close, binarization processing is performed with a plurality of pixel values (threshold values) to create a plurality of binary images. Since binarization targets all the pixels, if one side becomes n times, n ^ 2 times (n square times) pixels are searched.

  For this reason, when one side of the search range becomes longer, access to the pixels in the search range increases to a power of 2, and as a result, the processing time also increases to a power of 2. When pixel block extraction is performed by applying the MSER of Non-Patent Document 2 to the entire image, as shown in FIG. 5, access to the pixel increases to the power of the number of pixels on one side (long side) of the image. As a result, the analysis processing time is also multiplied by a power. It is assumed that the ratio of the long side to the short side of the image is 3 to 2. In the present embodiment, the relationship between the number of pixels on the long side of the image shown in FIG. 5 and the processing time is stored in advance in a memory or storage device for a predetermined ratio of the long side and the short side of the image, and a new image is processed. A processing time acquisition unit (not shown) that acquires the prediction time of the analysis processing from the number of pixels on the long side of In addition to the relationship shown in FIG. 5, in addition to the relationship between the number of images on the long side of the image and the processing time, the relationship between the number of pixels of the entire image and the processing time is prepared in advance and stored in a memory or storage device. Processing time can also be acquired.

  Next, a description will be given of the processing of the present embodiment that shortens the processing time while maintaining the extraction accuracy of the character region. In the present embodiment, a range where there is a high possibility that a character area exists is determined, and the image is reduced within the range at an appropriate reduction ratio to perform a character area extraction process. Since there is a possibility that a character area may exist in an image part other than the range where the character area is highly likely to exist, it is highly likely that a character area exists from the input image so that extraction of the character area is reduced. The character area is searched for the image part excluding the range. Alternatively, the character area is searched for an image portion excluding a range where the character area is likely to exist and an area where the character area is unlikely to exist from the input image.

  As described above, the pixel block generation by the analysis in units of pixels is performed, and the character region extraction processing is performed. The flow will be described with reference to the flowchart of FIG. The flowchart in FIG. 6 is for extracting an area of a race bib number as a character area, using an image obtained by imaging the marathon event or road race shown in FIG.

  Prior to the description of the flowchart of FIG. 6, the outline of the process in which the character area extraction unit 116 generates a pixel block by pixel-by-pixel analysis and extracts the character area in step S302 will be described in detail using the flowchart of FIG. 4. explain.

  A pixel block is generated from the image input in step S401. The generation of the pixel block can be realized by binarizing the image and extracting black pixels, for example. By using the MSER of Non-Patent Document 2, a pixel block with reduced noise can be extracted by extracting stable pixel blocks existing in a plurality of binary images binarized at a plurality of levels.

  However, since the pixel block other than the character is also included at this point, in step S402, a pixel block that is highly likely to be a character is determined from the pixel block obtained in step S401. Machine learning can be used as one form of determining whether or not a character pixel block. This is performed by using a discriminator that learns the feature quantity in advance so that the size or density of the pixel chunk is extracted as a feature quantity, and whether or not it is a character pixel chunk. Identification is performed on all pixel blocks obtained in step S401, and pixel blocks other than those determined as pixel blocks in the character area in step S402 are regarded as non-characters, and are excluded from processing targets in step S403.

  Next, in step S404, the pixel blocks selected as the character pixel blocks are collected using the positional relationship. For example, adjacent pixel blocks having an interval of 1/2 or less of the width of the pixel block are set as the same group. The value for determination may be obtained statistically from between adjacent pixel blocks. In addition, even in the pixel block excluded in step S403, if it exists between pixel blocks determined to be in the same group, the pixel block of the group may be used. As a result, a plurality of pixel blocks constituting a part of the character are grouped as one group and become character region candidates.

  In step S405, the background portion of the pixel block group obtained in step S404 is analyzed to determine whether or not the grouped pixel block is a pixel block in the same character area. Assuming that the background of the character area is almost uniform or gently changing, set a circumscribed quadrilateral of the pixel block, and average the luminance of the background that is the pixel block in the quadrangle and the other pixels A value, a variance value, and the like are obtained, and it is determined whether or not the obtained value satisfies the standard. The reference value is determined in advance by measuring an actual image. For example, the difference between the average luminance of the foreground and the background is 10 or more, or the luminance dispersion value of the background pixels is 70 or less.

  In step S406, the character region to be extracted is determined from the pixel block group and the background information obtained by the above processing. In this processing step, the character regions to be finally extracted are determined by further combining the grouped pixel blocks or deleting unnecessary pixel blocks. The processing method described above is an example of character region extraction processing, and is not limited to the method shown in FIG.

  This embodiment using the character region extraction processing method of step S302 described in FIG. 4 will be described in detail using the flowchart of FIG.

  In step S601, it is determined whether the number of pixels of the input image is equal to or less than a preset threshold value. If the image is smaller than the threshold, the processing time acquired by the acquisition unit (not shown) from within, for example, FIG. 5 is within a predetermined value, so that the extraction or reduction processing of the face region as the feature region is not performed, and in step S615 A region of a bib number that is a character region is extracted from the entire image. In order to acquire the processing time from the threshold value of the number of pixels of the input image, it is calculated in advance from the relationship between the processing time and the number of pixels (processing amount). For example, when the processing time is desired to be within T seconds of a predetermined value, the number of pixels on the long side that becomes the predetermined value T seconds of analysis processing time is acquired from the storage device that holds the values of the graph of FIG. Is a threshold value. If the number of pixels on the long side is equal to or less than Pt, the processing can be completed within a predetermined value T seconds even if the bib number area is extracted from the entire image.

  If an image exceeding the threshold value of the number of pixels is input in step S601, the face area detection unit 112 performs face detection in step S602. The face detection method can be implemented using conventional techniques. Depending on whether or not a face is detected in step S602, the flow of processing branches in step S603. Even if a face is not detected in step S602, there is a possibility that a race bib number is shown, so it is necessary to perform an analysis process of bib number number extraction.

  If no face area is detected as the feature area, in step S614, the image reduction unit 114 calculates a reduction ratio by a predetermined method, reduces the entire image, and then proceeds to step S615. The reduction ratio is calculated within a range in which the extraction accuracy does not decrease with respect to the size of the bib number (character area) to be extracted.

  First, the size of the bib number (character area) to be extracted may be determined directly or by the following method. The minimum occupancy ratio for the image of the participant who is the subject is determined in advance. Assuming that the aspect ratio of the image and the aspect ratio of the bib number are constant, the subject (participant) width of the subject (participant) is 20% or more as the occupation ratio of the subject in the image. Suppose you want to extract as an analysis target.

  If the length of the long side of the character area (bib number) is 40% on average with respect to the width of the subject, a character area in which the ratio of the long side to the long side of the image is 8% or more is extracted as an analysis target. Will do. Here, when an image having a long side of 3500 pixels is calculated as an example, a bib number having a long side of 280 pixels or more is extracted.

  Next, in order to extract the bib number, for example, if the extraction accuracy is ensured if the long side of the character area is 80 pixels or more, the minimum reduction rate within a range where the bib number extraction accuracy does not decrease is obtained. The reduction rate is about 29%. The reduction ratio varies depending on the size of the image to be processed. For example, if the number of pixels on the long side of the image is 4500 pixels, the reduction ratio is about 22% when calculated as described above. That is, the larger the image, the smaller the reduction rate, and the greater the effect of shortening the processing time.

  If a face is detected in step S602, the process proceeds to step S604. FIG. 7 shows a state in which face areas 701 and 702 are detected by the face area detection unit 112. It should be noted that the face area detection of the left hand participant has failed.

  In step S604, the search range determination unit 113 estimates the size and position of the character region including the bib number from the size and position of the detected face region, and determines the first search range. Since the average face size and number number are known information, the range where the number is likely to exist can be set from the position and size of the face region. For example, a width that is twice the width of the face area is set as the width of the range where the bib number exists, and a range from the bottom of the face area to a range that is three times the height of the face area is set as the first search range. When a plurality of face areas are detected, the first search range is determined for each face area. FIG. 7 shows how the first search ranges 703 and 704 are determined. In the first search range, the ratio of uniforms and skin is increased, and the noise of the pixel block is relatively small. Therefore, it is possible to shorten the time for the noise removal processing of the pixel block as compared with the region other than the first search range.

  After determining the first search range, a second search range that does not include the face region and the first search range is determined. There is a method in which a range obtained by masking the face region and the first search range from the input image is set as the second search range.

  Moreover, in order to further shorten the processing time, the first search range and the second search range can be made smaller. For example, the first search range can be made smaller by setting the range from the bottom of the face region to twice the width of the face region and twice the height of the face region as the first search range. On the other hand, from the input image, a range of twice the width of the face region and five times the height of the face region from the bottom of the face region and five times the height of the face region is masked to be a second search range. . Thereby, a 2nd search range can be made into a smaller range.

  That is, the size of the image to be processed is determined by more accurately identifying the first search range where a character region is likely to exist, and at the same time excluding the region where the character region is very unlikely to exist from the second search range. (Number of pixels) can be reduced.

  In step S605, the image reduction unit 114 determines the reduction ratios of the first search ranges 703 and 704, respectively. The size of the character area (bib number) that does not deteriorate the character extraction accuracy is acquired from the ROM 103 or the external storage device 105 in advance. The size of the character region (bib number) is estimated from the size of each face region detected in step S602, and the reduction ratio is determined from the size of the character region acquired in advance and the size of the character region estimated from the face region. . For example, if the width of the face area is W and the width of the character area is 50 pixels or more, if the character extraction accuracy does not decrease, the reduction ratio is 50 / W.

  By determining the reduction rate of the first search range in this way, reduction rates that do not reduce the character extraction accuracy can be obtained for the first search ranges 703 and 704 having different sizes. The first search ranges 703 and 704 can be reduced at the respective reduction ratios, and the character area can be extracted with a reduced processing time without lowering the accuracy of character area extraction.

  In addition, the bib number is extracted so that there is no leakage from the portion of the input image excluding the first search range and the face region, and the reduction rate is also applied to the second search range excluding the first search range and the face region from the input image. To decide. The reduction rate of the second search range can be determined with reference to the reduction rate of the first search range. For example, when only one face area is detected, the reduction rate of the second search range is set to the same value as the reduction rate of the first search range. When a plurality of face areas are detected, the same reduction ratio as the first search range determined from the largest face area may be used. Alternatively, the calculation may be performed using the reduction ratios of the plurality of first search ranges.

  Alternatively, by using the method described in step S614, a reduction rate that does not reduce the extraction accuracy is obtained from the size of the number number (character area) to be extracted, and is set as the reduction rate of the second search range. Also good.

  For example, when trying to extract a bib number with a long side number of 320 pixels or more, if the long side number of pixels required to extract the bib number is 80 pixels, the minimum number of the second search range The reduction rate is 25%. Since the character area is extracted after the second search range is reduced at the reduction ratio of the second search range, the number search number greater than or equal to the desired size in the second search range is extracted without reducing the extraction accuracy. Time can be shortened.

  As described above, the character area is extracted by reducing the character area independently with respect to the first search range and the second search range while suppressing the omission of extraction of the character area. Therefore, the processing time of the extraction process can be shortened compared to extracting the character area by reducing the entire image with the minimum reduction ratio when the entire image is reduced.

  In step S606, the image reduction unit 114 reduces the image in the first search range with the reduction rate determined in step S605 with respect to the first search range determined in step S604. In step S607, the image reduction unit 114 determines the number from the reduced partial image. Extract the number area. In step S608, it is determined whether the bib number area has been extracted from all the first search ranges. If the extraction process for all the first search ranges is completed, the process proceeds to step S609. FIG. 8 shows a reduced image in step S606. FIG. 8A is an original image of the first search range 703, and FIG. 8C is an image after reduction. Similarly, (b) is an original image of the first search range 704, and (d) is an image after reduction. The characters of the number numbers after the reduction are almost the same size within a range where the extraction accuracy does not decrease.

  In step S609, the second search range is reduced at the reduction rate determined in step S605. In step S610, the searched face area and the first search range are masked. FIG. 9A shows a state in which the reduced image is masked. That is, a range obtained by removing the face area and the first search range from the original image is the second search range.

  In step S611, the area of the bib number is extracted from the reduced image of the second search range without analyzing the masked range. In step S612, a determination unit (not shown) determines whether or not the number number area obtained from the second search range covers the mask range. For example, it is determined whether or not the character area (number number area) is in contact with the first search range and the ratio between the width and height (aspect ratio, width / height) of the character area is smaller than a predetermined value. The In this case, since only the partial area that is a part of the character area has been extracted, there is a high possibility that all the bib numbers cannot be extracted.

  If it is determined that the character area (number number area) falls within the mask range, a part of the extracted character area (number number area) is in the first search range, but cannot be extracted by mask processing. The letters (numbers) may be hidden. In step S613, the second search range is expanded, and the area extraction of the bib number is continued. For the direction of expansion, the direction of the character line is estimated from the already extracted characters, and the second search range is expanded toward the first search range in contact with the estimated direction of the character line. Alternatively, the number number of the participant may be regarded as being almost horizontal, and the second search range may be expanded toward the first search range in the horizontal direction. FIG. 9B shows a state in which the character area is expanded so that the aspect ratio of the character area is at least a predetermined value. As a result, it is possible to extract all the bib number areas on the image.

  In FIG. 6, it is described that the processing from step S <b> 606 to step S <b> 608 for the first search range is finished and then the processing after step S <b> 609 for the second search range is performed, but it is not necessary to limit to this. After the first search range and the second search range are determined, the process for the first search range and the process for the second search range may be performed in parallel. When the plurality of character region extraction units 116 perform processing in parallel using the plurality of CPUs 102, the processing time of the entire image can be significantly reduced.

  In addition, in FIG. 6, the method of extracting the character area by reducing the first search range and the second search range has been described, but according to the size of the first search range (image size) in the input image. Thus, the first search range may not be reduced. For example, when the size of the first search range in the input image is small as in the first search range 704 in FIG. 7, the reduction ratio is large (the change in the image size before and after the reduction process is small). Need not be reduced.

  As described above, it is possible to reduce the processing time by dividing the range in which the character region extraction process is performed on an image having a large number of pixels and reducing the image within a range in which the accuracy does not decrease. In addition, the search range is set so as not to divide the bib number area, which is a character area, as much as possible, the image is reduced within a range where accuracy is not reduced, and each divided image is reduced to greatly reduce the processing time. There is.

(Second Embodiment)
Hereinafter, a second embodiment will be described. Since the configuration diagram of the image processing apparatus according to the second embodiment is the same as that of the first embodiment, the description thereof will be omitted.

  A flow of processing for reading character information from an image, which is executed by the image processing apparatus in the present embodiment, will be described with reference to a flowchart of FIG. Note that description of the same processing steps as those in the first embodiment is omitted.

  In step S1001, it is determined whether the number of pixels of the input image is equal to or less than a preset threshold value. When the threshold value is exceeded, the processing steps from step S1002 to step S1005 are the same as in the first embodiment, the first search range and the second search range are determined, and the first search range and the second search that are divided into images are determined. A reduction ratio is determined for each partial image in the range.

  In step S1006, the amount of computation of the partial image reduction process and the extraction of the bib number area from the reduced partial image is performed both when the partial image is performed and when the entire image is not divided and is performed on the reduced image. Calculate and compare. If the amount of calculation when the image is divided and processed is small, that is, if there is an effect of reducing the processing time, the process proceeds to step S1007.

  The processing steps from step S1007 to step S1013 are the same as those in the first embodiment, and the bib number area is extracted from the first search range and the second search range.

  If it is determined in step S1006 that there is no effect of shortening the processing time, the process proceeds to step S1015, the reduction ratio is calculated by a predetermined method, and the entire input image is reduced. The reduction rate calculation method is the same as the calculation method in step S614 of FIG. 6 described in the first embodiment. In step S1016, a bib number area is extracted from the entire input image.

  Note that the processing in the case where the number of pixels of the input image is equal to or smaller than the threshold value in step S1001 or the case where the number range is applied to the mask range in step S1013 is the same as in the first embodiment.

  As described above, according to the present invention, it is possible to select a higher-speed processing step by performing processing with partial images after evaluating the processing time reduction effect.

(Third embodiment)
Although the above embodiment has been described using the example of face detection and bib number, it is not necessary to be limited to detecting a face region as a feature region. In the present embodiment, the first search range is determined by detecting the marker area instead of the face area as the feature area, and the bib number area that is the character area is detected from the first search range. The marker region can be detected by a conventional marker detection method using a marker region detection unit (not shown) instead of the face region detection unit 112.

  The marker is a specific pattern printed on the upper left corner of the bib as shown in FIG. 11 (a) or (b), but the position and shape of the marker need not be limited to this. The relative position between the marker position and the bib number need only be known in advance. Since the shape and size of the marker pattern are determined, the first search range that is likely to include the bib number is determined from the position and size of the marker. Then, after determining the first search range, the second search range is determined by the same method as in the first embodiment.

(Fourth embodiment)
Although the said embodiment demonstrated using the example which detects a number number, it is applicable also when detecting the registration number of the number plate of a motor vehicle. In this case, an automobile logo mark or an automobile headlight is detected as a feature area, and a first search range that is likely to include a license plate registration number is determined from the detected position and size of the feature area. Other processes are the same as in the above embodiment.

<Other embodiments>
The present invention supplies a program that realizes one or more functions of the above-described embodiments to a system or apparatus via a network or a storage medium, and one or more processors in the computer of the system or apparatus read and execute the program This process can be realized. It can also be realized by a circuit (for example, ASIC) that realizes one or more functions.

101 system bus 102 CPU
103 ROM
104 RAM
105 External storage device 106 Network interface 107 Display unit 108 Operation unit

Claims (14)

Detection means for detecting a feature region from the input image;
Determining means for determining a first search range in the input image and a second search range different from the first search range based on the position and size of the detected characteristic region;
Reduction means for reducing the second search range at a predetermined reduction rate;
Extracting means for extracting a character region using data for each pixel constituting each of the image of the first search range and the reduced image of the second search range subjected to the reduction process;
An image processing apparatus comprising:   The reduction means reduces the first search range, and the extraction means extracts a character area for each reduced image of the first search range and the second search range that have been reduced. The image processing apparatus according to claim 1.   Determining means for determining whether the aspect ratio of the character area extracted from the second search range is smaller than a predetermined value; and when the determining means determines that the aspect ratio of the character area is smaller than a predetermined value, 3. The image processing according to claim 1, further comprising: an extension unit that extends a second search range, wherein the extraction unit extracts a character region from the extended second search range. apparatus.   The extension means extends the second search range by extending the character region toward the first search range in contact with the character region in the direction of the character line of the extracted character region. The image processing apparatus according to claim 3.   The image processing apparatus according to claim 4, wherein the expansion unit expands the aspect ratio of the character area until at least the predetermined value is reached.   6. The determination unit according to claim 1, wherein the determination unit determines a second search range by excluding the first search range expanded by the feature region and a predetermined height or width from the input image. The image processing apparatus according to claim 1.   7. The image processing according to claim 1, wherein the extraction unit extracts a character region in parallel from each of the reduced first search range and the second search range. apparatus.   The reduction ratio of the reduction processing of the first search range by the reduction means is calculated based on the size of the feature area detected by the detection means and the size of the character area that can be extracted by the extraction means. The image processing apparatus according to claim 1, wherein the image processing apparatus is an image processing apparatus.   The reduction ratio of the reduction process of the second search range by the second reduction unit is calculated based on the size of the character region extracted by the extraction unit and the size of the character region that can be extracted by the extraction unit. The image processing apparatus according to claim 1, wherein the image processing apparatus is an image processing apparatus.   The image processing apparatus further includes an acquisition unit that acquires a processing time for extracting a character region for the entire input image from the number of pixels of the input image, and if the processing time acquired by the acquisition unit is within a predetermined value, The image processing apparatus according to claim 1, wherein the extraction unit performs a process of extracting a character region from the entire input image.   The image processing apparatus according to claim 1, wherein the feature area is a face area.   The image processing apparatus according to claim 1, wherein the feature region is a marker region. A detection step of detecting a feature region from the input image;
A determination step of determining a first search range in the input image and a second search range different from the first search range based on the detected position and size of the feature region;
A reduction step of reducing the second search range at a predetermined reduction rate;
An extraction step of extracting a character region using data for each pixel constituting each of the image of the first search range and the reduced image of the second search range subjected to the reduction process;
An image processing method comprising: A detection step of detecting a feature region from the input image;
A determination step for determining a first search range in the input image and a second search range different from the first search range based on the detected position and size of the feature region;
A reduction step of reducing the second search range at a predetermined reduction rate;
An extraction step of extracting a character region using data for each pixel constituting each of the image of the first search range and the reduced image of the second search range subjected to the reduction process;
A program that causes a computer to execute.

Images