JP2010186246A - Image processing apparatus, method, and program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an image processing apparatus capable of accurately extracting the character area of a reversed character included in an input image, without using line information. <P>SOLUTION: An edge detection means 11 detects edges from an input image according to changes in feature quantity between adjacent pixels. An area extraction and separation means 12 separates pixels near the pixels of the input image corresponding to the detected edges into pixels with high feature quantities and pixels with low feature quantities according to the feature quantity. The area extraction and separation means 12 extracts as character regions pixels with low feature quantities that are surrounded by the pixels with the high feature quantities and pixels with high feature quantities that are surrounded by the pixels with the low feature quantities. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an image processing apparatus, method, and program, and more particularly to an image processing apparatus, method, and program for extracting a character region from an image.

  There is an optical character reader (OCR) that optically reads characters included in an image using a scanner device or the like and outputs character data. In order to obtain character data by OCR, it is necessary to classify a background region and a character region from an image and extract a region constituting a character. For example, in a black and white image in which characters are written in black on a white background, a white region is extracted as a background region, and a black region is extracted as a character region, thereby extracting the character region.

  By the way, various characters are often decorated in images such as advertisements and posters. For example, in an advertisement or the like, in many cases, a specific character is displayed as an inverted character in order to display the character more emphasized than surrounding characters. An inverted character refers to, for example, a character written in white characters on a black background when other characters are written in black on a white background. That is, the color which comprises a character points out the character reversed with respect to the circumference | surroundings. For example, in a surrounding character, white is a background and black is a character, and in an inverted character, black is a background and white is a character.

  When reverse characters are mixed in the image, the background region is erroneously extracted as the character region unless the reverse region is accurately detected. Regarding detection of inverted characters, there are techniques described in Patent Documents 1 to 4. In Patent Document 1, the run length is compared with a predetermined length set in advance, and a reverse image determination is performed based on whether or not the change in the run length satisfies a predetermined condition. The run length is defined by a length of pixels having the same density measured by measuring image information in one direction. In Patent Document 1, a histogram of the number of character edges and the number of black runs is generated for the extracted line area. Thereafter, it is determined whether or not the value of the histogram exceeds a predetermined threshold value, and it is determined whether or not it is an inversion region.

  In Patent Document 2, image inversion is determined based on the ratio of black pixels and white pixels in an input binary image. In Patent Document 2, the number of black pixels and white pixels in an input binary image is counted, and when the ratio of black pixels exceeds, for example, 70%, it is determined as an inverted image. Similarly in Patent Document 3, the number of black pixels and white pixels is counted, and a color having a large number of pixels is used as a background region.

  In Patent Document 4, an input image is binarized to obtain a binary image, and an edge image is generated by detecting an edge of the input image. Region identification processing is performed on each of the binary image and the edge image, and the region identification results of both are integrated. Thereafter, partial binarization processing of the input image is performed in accordance with each character area included in the integrated area identification result. In partial binarization, similar to Patent Document 2 and Patent Document 3, it is determined whether the character is an inverted character or a non-inverted character according to the ratio of white pixels to black pixels.

JP 2004-326568 A JP 2004-64664 A Japanese Patent No. 2743378 JP 2007-183742 A

Otsu “Automatic Threshold Selection Method Based on Discrimination and Least Squares Criterion”, IEICE Transactions, vol. J63-D, no4, pp. 349-356, 1980

  In Patent Document 1, on the assumption that line information such as the number of lines and the position of each line is known, the inversion area is specified using the number of runs of white pixels and black pixels for each line. In Patent Document 1, it is necessary to know in advance where a row exists in an image. Therefore, Patent Document 1 can be applied only to a standard document in which the position of a line is fixed. Further, the characters included in the input image are not necessarily arranged in a line vertically or horizontally. For example, characters may be arranged in an oblique direction, or characters may be arranged on a curved curve. This tendency is particularly strong in images with many decorations such as advertisements. Since Patent Document 1 performs processing for each line at a fixed position, a character region cannot be extracted from such an image.

  In Patent Document 2, it is determined whether or not the region is an inversion region based on the ratio of white pixels and black pixels in a circumscribed rectangle circumscribing the character region. In Patent Document 2, if the rectangular area is not properly taken, pixels in the background area that are not related to the character greatly affect the determination criterion, and the determination of the reversed character may be erroneous. For this reason, inverted characters cannot be extracted with high accuracy. Similarly, Patent Documents 3 and 4 also extract the inversion area from the ratio of black and white pixels in the circumscribed rectangle. For this reason, there is a case where reverse character extraction is erroneous depending on the number of characters in the rectangle and the typeface of the characters.

  An object of the present invention is to provide an image processing apparatus, method, and program capable of accurately extracting a character region of inverted characters included in an input image without using line information.

  In order to achieve the above object, an image processing apparatus of the present invention corresponds to an edge detection unit that detects an edge from an input image based on a change in a feature amount between adjacent pixels, and corresponds to the detected edge. A pixel in the vicinity of the pixel of the input image is divided into a pixel having a high feature amount and a pixel having a low feature amount based on the feature amount, and a pixel having a low feature amount surrounded by pixels having a high feature amount, and And an area extracting / separating means for extracting a pixel having a high feature quantity surrounded by pixels having a low feature quantity as a character area.

  The image processing method of the present invention includes a step in which a computer detects an edge from an input image based on a change in a feature amount between adjacent pixels, and the computer inputs the input corresponding to the detected edge. Dividing a pixel in the vicinity of an image pixel into a pixel having a high feature value and a pixel having a low feature value based on the feature value; and the computer has a low feature value surrounded by the pixels having a high feature value. Extracting a pixel having a high feature amount surrounded by pixels and a pixel having a low feature amount as a character region.

  The program according to the present invention allows a computer to detect an edge from an input image based on a feature amount change between adjacent pixels, and to detect the vicinity of the pixel of the input image corresponding to the detected edge. Based on the feature amount, the pixel is classified into a pixel having a high feature amount and a pixel having a low feature amount, a pixel surrounded by the pixels having a high feature amount, and a pixel having a low feature amount and a pixel having a low feature amount. It is characterized by executing a process of extracting a surrounded pixel having a high feature amount as a character region.

  The image processing apparatus, method, and program of the present invention can accurately extract the character region of the inverted character included in the input image without using line information.

1 is a block diagram showing an image processing apparatus of the present invention. 1 is a block diagram showing an image processing apparatus according to a first embodiment of the present invention. The flowchart which shows an operation | movement procedure. The figure which shows an input image. (a) And (b) is a figure which shows edge image A and B comprised by the 1st edge and the 2nd edge, respectively. (a) is a figure which expands and shows a part of input image, (b) and (c) is a figure which expands and shows a part of edge image A and the edge image B. (a) And (b) is a figure which shows the 1st image and the 2nd image, respectively. (a) And (b) is a figure which expands and shows a part of 1st image and 2nd image, respectively. (a) And (b) is a figure which shows the character area extracted from the 1st image and the character area extracted from the 2nd image, respectively. (a) And (b) is a figure which shows the character area extracted from Fig.8 (a) and (b), respectively. (a) And (b) is a figure which shows the label area | region in a 1st image and a 2nd image, respectively, (c) is a figure which shows the character area after integration. The figure which shows the image processing apparatus of 2nd Embodiment of this invention. The flowchart which shows the operation | movement procedure in 2nd Embodiment. (a) is a figure which shows a part of reduced image, (b) is a figure which shows the area | region corresponding to an original image.

  First, an outline of the present invention will be described. FIG. 1 shows an image processing apparatus of the present invention. The image processing apparatus 10 includes an edge detection unit 11 and a region extraction / separation unit 12. The edge detection unit 11 detects an edge from the input image based on a change in feature amount between adjacent pixels. The region extraction / separation unit 12 classifies the pixels in the vicinity of the pixel of the input image corresponding to the edge detected by the edge detection unit 11 into a pixel having a high feature amount and a pixel having a low feature amount. The region extraction / separation unit 12 extracts, as a character region, pixels having a low feature amount surrounded by pixels having a high feature amount and pixels having a high feature amount surrounded by pixels having a low feature amount in the vicinity of the edge.

  The edge detected by the edge detection means 11 indicates the boundary between the character area and the background area. By extracting pixels with low feature values that are surrounded by pixels with high feature values in the vicinity of the edge, the pixels with the high feature values are used as the background, and the character regions of the characters in which the low feature values make up the character Can be extracted. In addition, by extracting pixels with a high feature value that are surrounded by pixels with a low feature value in the vicinity of the edge, a character having a low feature value as a background and a pixel with a high feature value constituting a character Regions can be extracted. That is, it is possible to extract character regions of both inverted characters and non-inverted characters from the input image.

  In the present invention, an edge is detected from an input image, and a character region is extracted from the vicinity of the edge. For this reason, it is not necessary to know line information in advance when extracting a character area of an inverted character. Therefore, character regions can be extracted from various images as well as standard document images. In addition, since the character area is extracted according to the feature amount of the pixel near the edge, the character area is discriminated in advance, and the input image is binarized in the circumscribed rectangle circumscribing the character area. There is no need to obtain the ratio of white pixels to black pixels. When determining the ratio of the white pixel to the black pixel in the circumscribed rectangle and determining whether or not it is an inversion region, the background pixel may affect the determination of the inversion region. In some cases, the inversion area cannot be correctly determined according to the number of characters in the circumscribed rectangle or the typeface of the characters. In the present invention, such a problem does not occur, and the character region of the reversed character can be extracted with high accuracy.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. FIG. 2 shows an image processing apparatus according to the first embodiment of the present invention. The image processing apparatus 100 includes an image input device 110, a data processing unit 120, a data storage unit 130, and an image output device 140. The image input device 110 is a device for inputting an image. The image input device 110 is typically an imaging system such as a still camera, a video camera, or a scanner. The image input device 110 inputs image data to the data processing unit 120. Each pixel of the input image has a pixel value. In the case of a monochrome gradation image, the pixel value is represented by a luminance value from black (lowest luminance) to white (highest luminance). If the input image is a color image, the pixel value is represented by a vector value corresponding to the color space.

  The data processing unit 120 operates by operating a predetermined program on the electronic computer. The data processing unit 120 includes an edge detection unit 121, a region extraction / separation unit 122, and a character region integration unit 125. The edge detection unit 121 detects an edge from the input image based on a change in feature amount between adjacent pixels. The edge detection unit 121 obtains a change in the feature amount between adjacent pixels, and detects a pixel whose feature amount change (edge strength) is equal to or greater than a threshold as an edge.

  The edge detection unit 121 detects the first edge and the second edge according to the direction of change of the feature amount. The edge detection unit 121 detects an edge in the direction in which the feature amount increases as the first edge, and detects an edge in the direction in which the feature amount decreases as the second edge. In other words, the edge detection unit 121 detects, as a first edge, a pixel having a lower feature value among two adjacent pixels having a large change in the feature value, and detects a pixel having a higher feature value as the second edge. Detect as an edge.

  The edge detection unit 121 performs edge detection for the luminance component of the input image, for example. The edge detection unit 121 detects, from the input image, a pixel in which a change in luminance value between adjacent pixels is equal to or greater than a threshold value and a pixel that changes from a bright luminance value to a dark luminance value as a first edge. Further, the edge detection unit 121 detects, as the second edge, a pixel in which the change in luminance value between adjacent pixels is equal to or greater than a threshold value and changes from a dark luminance value to a bright luminance value. When the input image is a color image and the pixel value of each pixel is represented by a luminance value of RGB, the edge detection unit 121 performs edge detection processing on each RGB plane, and the edge intensity at the same pixel position of the three planes Of these, the edge may be detected by thresholding the maximum edge strength.

  When the input image is a color image, the edge detection unit 121 may detect the edge based on these changes using hue and saturation as the feature amount. For example, the edge detection unit 121 detects, as a first edge, a pixel in which the hue change of an adjacent pixel is equal to or greater than a threshold value and the saturation changes from a large value to a small value from the input image. Further, the edge detection unit 121 detects, as the second edge, a pixel in which the hue change of an adjacent pixel is equal to or greater than a threshold value and the saturation value changes from a large value to a small value. In the following, for simplification of description, a case where the input image is a monochrome gradation image will be mainly described.

  The region extraction / separation unit 122 receives edge position information from the edge detection unit 121. The region extraction / separation unit 122 classifies the pixels of the input image near the first edge into pixels having a higher feature amount than the surroundings and pixels having a lower feature amount than the surroundings based on the feature amount. For example, the region extraction / separation unit 122 performs threshold processing on the feature amount, divides pixels having a feature amount larger than the threshold value into pixels having a higher feature amount, and determines pixels having feature amounts smaller than the threshold as feature amounts Divide into low pixels. The region extraction / separation unit 122 includes, among the pixels in the vicinity of the pixel of the input image corresponding to the position of the first edge, pixels surrounded by pixels having a higher feature amount than the surroundings and having a lower feature amount than the surroundings. Extract as a character area.

  In addition, the region extraction / separation unit 122 includes a pixel having a feature amount lower than the surroundings and a feature amount lower than the surroundings of pixels in the vicinity of the pixel of the input image corresponding to the position of the second edge detected by the edge detection unit 121. Is divided into high pixels. Similar to the above, this classification can be performed by thresholding the feature amount. The region extraction / separation unit 122 includes a pixel having a feature amount higher than that of the surrounding pixels surrounded by pixels having a feature amount lower than that of the pixels in the vicinity of the pixel of the input image corresponding to the position of the second edge. Extract as a character area. Note that “enclosed” here does not need to be completely surrounded, and is sufficient if the periphery is covered.

  In the present embodiment, in the input image, a pixel having a low feature value is used as a background, and a character that is composed of a pixel having a high feature value is used as an inverted character. Conversely, in the input image, a pixel having a high feature value is set as a background, and a character composed of pixels having a low feature value is set as a non-inverted character. A character region composed of pixels having a feature amount lower than that of the surrounding pixels among pixels near the pixel corresponding to the position of the first edge is a non-inverted character. Corresponds to the character area. In addition, a character area composed of pixels having a feature amount higher than that of the surrounding pixels among the pixels in the vicinity of the pixel corresponding to the position of the second edge is a reversed character. Corresponds to the character area.

  The area extraction / separation means 122 includes a local binarization means 123 and a character area separation means 124. The local binarization unit 123 binarizes the input image in the vicinity of the first edge with the pixel near the pixel of the input image corresponding to the first edge detected by the edge detection unit 121 as a processing target. The local binarization unit 123 binarizes the input image in the vicinity of the second edge with the pixel near the pixel of the input image corresponding to the second edge detected by the edge detection unit 121 as a processing target. .

  The local binarization unit 123 sets a pixel having a low feature value as black and a pixel having a high feature value as white in the pixels near the first edge. This image is the first image. Further, the local binarization unit 123 sets pixels having a higher feature amount than the surrounding pixels to black and pixels having a lower feature amount to white in the pixels near the second edge. This image is the second image. The white pixel and the black pixel (the role of the white pixel and the black pixel) may be reversed.

  The character area separating unit 124 extracts black pixels surrounded by white pixels from the first image and outputs them as character areas. The character area extracted from the first image corresponds to a character area composed of pixels surrounded by pixels having a high feature amount and having a low feature amount, that is, a character region of non-inverted characters. In addition, the character area separation unit 124 extracts black pixels surrounded by white pixels in the second image and outputs them as character areas. The character region extracted from the second image corresponds to a character region composed of pixels with a high feature amount surrounded by pixels with a low feature amount, that is, a character region of inverted characters.

  Here, when a background enters a gap between character parts in a non-inverted character, the portion appears as a black pixel surrounded by white pixels in the second image. In addition, when a background enters a gap between character parts in an inverted character, the portion appears as a black pixel surrounded by white pixels in the first image. These become noise components in the extracted character area. Therefore, the noise component is removed assuming that the characters constituting one character are either inverted characters or non-inverted characters.

  The character area integration unit 125 integrates the character area extracted from the first image and the character area extracted from the second image. At the time of integration, when the character area extracted from the first image and the character area extracted from the second image overlap each other, the character area integration unit 125 selects any one of the character areas at the overlapping positions. One is selected as a character area. More specifically, the character region integration unit 125 performs a labeling process on the character region extracted from the first image and the character region extracted from the second image. The character region integration unit 125 selects one of the labels at the overlapping positions based on the label area.

  The image output device 140 outputs the character region that has been subjected to the integration processing by the character region integration unit 125 to a subsequent character recognition device or the like. Alternatively, the image output device 140 may display the character area on a display screen such as a display. The image output device 140 may separately output the character area extracted from the first image and the character area extracted from the second image. At that time, the image output apparatus 140 may delete, from the first image or the second image, the character area that is not selected by the character area integration unit 125 from the overlapping character areas. The data storage unit 130 includes a parameter storage unit 131. The parameter storage unit 131 stores various threshold values, parameters for determining the binarization processing method by the local binarization unit 123, and the like.

  FIG. 3 shows an operation procedure. The image input device 110 inputs an image captured by the imaging device to the data processing unit 120 (step S1). The input image is not limited to the captured image, and may be an image generated by some method. The edge detection unit 121 detects the first edge and the second edge from the input image (steps S2A and S2B). For edge detection, the edge detection unit 121 applies an edge detection filter such as a Laplacian filter to detect a position where the feature amount changes (a position where the sign is inverted). The edge detection unit 121 detects, as a first edge, a pixel having a low feature amount at a position where the feature amount changes, that is, a boundary between a pixel having a high feature amount and a pixel having a low feature amount, as a first edge. The pixel having the higher side is detected as the second edge.

  The edge detection unit 121 calculates the edge strength of each pixel and generates a binary edge image by thresholding the edge strength. A fixed value stored in the parameter storage unit 131 can be used as the threshold value of the edge strength. Or you may change a threshold value for every small area using a discriminant analysis method (nonpatent literature 1). Note that the Laplacian filter is susceptible to noise. Therefore, before the edge detection unit 121 performs edge detection, a median filter, a Gaussian filter, or the like may be applied in advance to perform noise removal processing or smoothing processing on the input image. When such processing is performed, erroneous detection of edges due to noise can be suppressed.

  The local binarization unit 123 receives the positions (coordinates) of the first edge and the second edge from the edge detection unit 121, and performs local binarization processing on pixels near the edge of the input image. (Steps S3A, S3B). In step S3A, for example, the local binarization unit 123 sets the pixels of the input image within a predetermined range from the pixels constituting the first edge as pixels near the edge, and binarizes the pixels near the edge. I do. The local binarization unit 123 compares the feature amount of each pixel in the processing target region with the feature amount of the surrounding pixels, and whites a pixel having a higher feature amount than the surrounding pixels and a pixel having a low feature amount. A first image is generated in black.

  In step S3B, the local binarization unit 123 sets pixels within a predetermined range from the pixels constituting the second edge as pixels near the edge, and binarizes the pixels near the edge in the same manner as described above. I do. The local binarization unit 123 compares the feature amount of the target pixel with the feature amount of the peripheral pixel for each pixel in the processing target region, and sets the pixel whose feature amount is lower than the peripheral pixel to white. A second image is generated in which pixels having a high value are black. The local binarization unit 123 performs threshold processing on the feature amount of the pixel to be processed using the fixed value stored in the parameter storage unit 131 when generating the first image and the second image, You may binarize. Moreover, the local binarization means 123 may determine a threshold value dynamically for every small area using a discriminant analysis method (nonpatent literature 1).

  The character area separating unit 124 extracts black pixels surrounded by white pixels from the first and second images, and uses them as character areas (steps S4A and S4B). Extraction of black pixels surrounded by white pixels can be performed by the following procedure. The character area separation unit 124 searches for connected white pixels in the first image. The character area separation unit 124 determines whether or not there is a black pixel among the connected white pixels in the first image. When the black pixel area is included in the white pixel area, the character area separating unit 124 extracts the black pixel area as the character area. The character region separation unit 124 extracts, as a character region, a black pixel region included in a connected white pixel region in the second image by the same procedure as described above.

  The character region integration unit 125 performs a labeling process on the character region extracted from the first image and the character region extracted from the second image. At the time of labeling, pixel expansion processing may be performed before labeling so that a plurality of parts constituting a separated character or a series of character strings become one label. The character region integration unit 125 compares the label areas of the regions where the label positions overlap among the character regions extracted from both images (step S5). The character region integration means 125 leaves the character region with the larger label area, erases the character region with the smaller label area, and integrates the character regions extracted from both images (step S6). By performing this integration process, in a certain label image, the character area extracted from either the first image or the second image remains as a character area after integration.

  Hereinafter, a specific example will be described. FIG. 4 shows an input image example. The input image includes white characters (inverted characters) composed of pixels with higher brightness than the background and black characters (non-inverted characters) composed of pixels with lower brightness than the background. The edge detection means 121 detects the first edge and the second edge from the input image shown in FIG. FIG. 5A shows an edge image (edge image A) composed of the first edge, and FIG. 5B shows an edge image (edge image B) composed of the second edge. If an edge is detected from the input image shown in FIG. 4 and the pixel at the edge position is black and the other pixels are white, the edge image is as shown in FIGS. 5 (a) and 5 (b).

  FIG. 6A shows an enlarged part of the input image, and FIGS. 6B and 6C show the edge image in an enlarged manner. The image shown in FIG. 6A corresponds to a portion of a point of “luck” characters. The dot portion has a high brightness, that is, a color close to white, and the background portion is gray, that is, a color of intermediate gradation. The edge detection unit 121 detects a pixel whose luminance changes in the increasing direction as the first edge. That is, the edge detection unit 121 detects the position of the dark pixel as the first edge at the boundary between the dark pixel and the bright pixel in the image shown in FIG. 6A (FIG. 6B). Further, the edge detection unit 121 detects a pixel whose luminance changes in a decreasing direction as the second edge. That is, the edge detection unit 121 detects the position of the bright pixel as the second edge at the boundary between the dark pixel and the bright pixel in the image shown in FIG. 6A (FIG. 6C). The first edge and the second edge are adjacent to each other.

  The local binarization means 123 uses the position of the first edge to make black and white 2 pixels of the input image in the vicinity of the first edge where the low luminance pixel is black and the high luminance pixel is white. A value image (first image) is generated. Further, the local binarization unit 123 uses the position of the second edge to make black and white a pixel having a high luminance and white a pixel having a low luminance among pixels of the input image near the second edge. The binary image (second image) is generated. 7A and 7B show the first image and the second image. In FIGS. 7A and 7B, a gray area is an area not subject to binarization processing. This area corresponds to the background area.

  FIGS. 8A and 8B are enlarged views of a part of the first image and the second image, respectively. The portion shown in FIG. 8 corresponds to the portion of the silver point shown in FIG. The local binarization means 123 performs binarization processing in the vicinity of the edge position (black pixel position) shown in FIGS. The range of pixels in the vicinity of the edge is defined as a 3 × 3 area that is vertically and horizontally, and diagonally adjacent to the edge position. The pixel range in the vicinity of the edge may be determined as appropriate in consideration of the fineness of the lines constituting the character, the size of the character in the input image, the resolution of the input image, and the like.

  In the input image (FIG. 6A), the inner portion (the portion that appears white) of the first edge has higher brightness than the surroundings, and the first edge and the outer portion (the portion that appears gray). The brightness is lower than the surroundings. The local binarization unit 123 sets the portion with high luminance, that is, the portion inside the first edge as white, and the portion with low luminance, that is, the first edge and the outside portion as black. 1 image is generated (FIG. 8A). In the input image, the brightness of the second edge and the portion inside the second edge is higher than that of the surroundings, and the brightness of the portion outside the second edge is lower than that of the surroundings. The local binarization unit 123 sets the high-luminance portion, that is, the second edge and the inner portion thereof as black, and the low-luminance portion, ie, the portion outside the second edge as white. 2 image is generated (FIG. 8B).

  The character area separating unit 124 extracts black pixels surrounded by white pixels from the first image and the second image, and extracts a character area from both images. 9A and 9B show the character areas extracted by the character area separating unit 124. FIG. In the first image (binary image in the vicinity of the first edge) shown in FIG. 7A, when black pixels surrounded by white pixels are left and black pixels not surrounded by white pixels are deleted, FIG. The image shown in 9 (a) is obtained. A black pixel in the image shown in FIG. 9A corresponds to a character region extracted from the first image. Further, in the second image (binary image in the vicinity of the second edge) shown in FIG. 7B, when black pixels surrounded by white pixels are left and black pixels not surrounded by white pixels are deleted. An image shown in FIG. 9B is obtained. A black pixel in the image shown in FIG. 9B corresponds to a character region extracted from the second image.

  FIGS. 10A and 10B show character areas extracted from the first image and the second image shown in FIGS. 8A and 8B, respectively. The character region separation unit 124 searches for white pixels (a lump of white pixels) that are continuous vertically, horizontally, and diagonally in the first image shown in FIG. When the white pixel block is found, the character area separating unit 124 checks whether there is a black pixel in the white pixel block. In the input image (FIG. 4), the character “luck” is an inverted character, and the point portion of “shinyu” is a high luminance region surrounded by a low luminance region. There are no black pixels in the block of pixels. Therefore, there is no character region extracted from the first image shown in FIG. 8A (FIG. 10A).

  The character region separation unit 124 searches for a cluster of white pixels that are continuous vertically, horizontally, and diagonally in the second image shown in FIG. When the white pixel block is found, the character area separating unit 124 checks whether there is a black pixel in the white pixel block. In FIG. 8B, since the white pixels surround the black pixels, the black pixels existing in the white pixel block are found. The character area separating unit 124 extracts black pixels surrounded by white pixels in FIG. 8B as character areas (FIG. 10B). Thus, the character region corresponding to the point portion of “shinnyu” is not extracted from the first image but extracted from the second image.

  The non-inverted character has high luminance of pixels constituting the background and low luminance of pixels constituting the character. Considering the vicinity of the edge of the non-inverted character, in the first image, the pixels constituting the non-inverted character in the vicinity of the first edge are black pixels, and the background pixel of the non-inverted character is the white pixel. Further, in the second image, the pixels constituting the non-inverted character near the second edge are white pixels, and the background pixel of the non-inverted character is a black pixel. Since the non-inverted characters are composed of pixels with low luminance surrounded by pixels with high luminance, by extracting black pixels surrounded with white pixels, the character region of the non-inverted characters is extracted from the first image. Can be extracted.

  In the inverted character, the luminance of the pixels constituting the background is low, and the luminance of the pixels constituting the character is high. When the vicinity of the edge of the inverted character is considered, in the first image, the pixels constituting the inverted character near the first edge are white pixels, and the background pixel of the inverted character is a black pixel. In the second image, the pixels constituting the reversed character near the second edge are black pixels, and the background pixels of the reversed character are white pixels. Since the reverse character is composed of pixels with high luminance surrounded by pixels with low luminance, the character region of the reverse character can be extracted from the second image by extracting the black pixels surrounded by white pixels. . That is, the character area of non-inverted characters can be extracted from the first image, and the character area of inverted characters can be extracted from the second image.

  The character region extracted from the first image includes pixels in the background portion surrounded by pixels that form characters with reversed characters, in addition to pixels that form non-inverted characters. For example, in the “for” character that is an inverted character, the background pixel inside the character is surrounded by the pixels constituting the character (FIG. 4). Inverted characters have high brightness in the pixels constituting the text and low brightness in the background pixels. Therefore, the background pixels inside the “use” characters are black pixels surrounded by white pixels in the first image. It appears and is extracted as a character area (FIG. 9A). Similarly, in the second image, the background pixel inside the non-inverted character is extracted as a character region (FIG. 9B). These background pixels extracted as the character area become noise components.

  The character region integration means 125 assumes that, at the same position, the character is composed of either an inverted character or a non-inverted character, and the character region integrating means 125 is located inside the inverted character included in the character region extracted from the first image. The background pixel and the background pixel inside the non-inverted character included in the character region extracted from the second image are removed. The character area integration unit 125 integrates both character areas from which background pixels are removed. In other words, when the character areas extracted from both images are at overlapping positions, the character area integration unit 125 calculates the character area extracted from the first image and the character area extracted from the second image. Either one is selected as a character area after integration.

  FIGS. 11A to 11C show the character region integration process. The character region integration unit 125 performs pixel expansion processing on the character region extracted from the first image, and labels the portion where the black pixels are connected (FIG. 11A). Similarly, the character region integration unit 125 performs pixel expansion processing and labeling on the character region extracted from the second image (FIG. 11B). By performing pixel dilation processing before labeling, the character region extracted from the second image (FIG. 9B) is a character region of “advanced function” that has been separated character by character. Are connected to adjacent characters, and the same label is assigned to the character string.

  The character area integration unit 125 checks whether or not there is an overlapping label between the character area label extracted from the first image and the character area label extracted from the second image. . When there is no label at an overlapping position in both character areas, that is, at a position where the label exists only in one of the character areas, the character area integration means 125 (the first image or A black pixel extracted from the second image) is set as a character region after integration. When there is a label at an overlapping position, the character region integration unit 125 compares the size (area) of the label region, and the character region with the larger label area is set as the character region after integration (FIG. 11). (C)). Instead of the label area, the number of pixels in the label may be compared.

  For example, consider the “advanced function” part included in the input image (FIG. 4). Since this part is an inverted character, the character region is extracted from the second image (FIG. 9B). However, from the first image, pixels with high luminance surrounded by pixels with low luminance are extracted as character regions (FIG. 9A). When the character area integration unit 125 performs labeling, six labels are generated in the character area extracted from the first image (FIG. 11A), and one label is generated in the character area extracted from the second image. (FIG. 11B). The character area integrating unit 125 compares each of the six labels existing in FIG. 11A with the labels existing in FIG. The label areas of the six labels existing in FIG. 11A are all smaller than the label areas of the labels existing in FIG. Therefore, the character area integration unit 125 selects the character string extracted from the second image for the “advanced function” portion. Finally, the character region extracted from the second image remains in the character region after integration (FIG. 11C).

  In the present embodiment, the edge detection unit 121 detects an edge from an input image based on a change in feature amount between adjacent pixels. The region extraction / separation unit 122 classifies the pixels in the vicinity of the pixel of the input image corresponding to the detected edge into a pixel having a high feature amount and a pixel having a low feature amount. The region extraction / separation unit 122 extracts, as a character region, pixels having a low feature amount surrounded by pixels having a high feature amount and pixels having a high feature amount surrounded by pixels having a low feature amount. The boundary between the character and the background is detected as an edge, and pixels near the edge include a pixel with a high feature value and a pixel with a low feature value. By extracting a pixel having a low feature amount surrounded by pixels having a high feature amount from pixels in the vicinity of the edge, a character region of a non-inverted character can be extracted. In addition, by extracting a pixel having a high feature amount surrounded by pixels having a low feature amount from pixels in the vicinity of the edge, a character region of an inverted character can be extracted.

  In this embodiment, since both the character area of the inverted character and the character area of the non-inverted character are extracted from the vicinity of the edge, information on how the line is configured in advance when extracting the character area is obtained. There is no need to use it. Therefore, the input image is not limited to a standard document image, and character regions of inverted characters and non-inverted characters can be extracted from various input images. In addition, even when characters are arranged obliquely or arranged on a curved curve, the character regions of inverted characters and non-inverted characters can be extracted.

  Further, in the present embodiment, pixels having a high feature amount surrounded by pixels having a low feature amount in the vicinity of the edge are used as the character region of the reversed character. There is no need to set and calculate the ratio of black pixels to white pixels in the circumscribed rectangle. In the method of judging the inversion area based on the ratio of black pixels to white pixels, the inversion characters cannot be judged correctly depending on how to draw the circumscribed rectangle, the font of the characters, the thickness of the lines constituting the characters, etc. There is. On the other hand, in the present embodiment, a circumscribed rectangle is not necessary, and the character area of the inverted character can be extracted without depending on the font or line thickness. Therefore, the character region of the reversed character can be extracted with higher accuracy than the method of determining the reversed region based on the ratio between the black pixel and the white pixel.

  Subsequently, a second embodiment of the present invention will be described. FIG. 12 shows an image processing apparatus according to the second embodiment of the present invention. The image processing apparatus 100A includes an image input device 110, a data processing unit 120A, a data storage unit 130, and an image output device 140. The data processing unit 120A includes an edge detection unit 121, a region extraction / separation unit 122, a character region integration unit 125, an image reduction unit 126, and a region matching unit 127. The configuration of the data processing unit 120A in the second embodiment is a configuration in which an image reduction unit 126 and a region matching unit 127 are added to the configuration of the data processing unit 120 of the first embodiment shown in FIG.

  The image reduction unit 126 reduces the image input by the image input device 110 and passes it to the edge detection unit 121. The image reducing unit 126 reduces the input image to an image with a desired resolution. The image reduction unit 126 reduces the input image to a desired resolution using a known reduction method such as a bilinear method or a bicubic method.

  The image reduction rate is determined as follows, for example. A table is stored in the parameter storage unit 131 in advance. The table stores the correspondence between the character size of the character desired to be extracted from the image and the image resolution that enables extraction of the character area of the character size. The user specifies the character size to be processed. The image reduction unit 126 refers to the table to obtain a resolution corresponding to the character size designated by the user. The image reduction means 126 determines the image reduction ratio from the resolution of the input image and the resolution stored in the table. Alternatively, the correspondence between the character size to be processed and the reduction rate may be stored in the table. In this case, the image reduction unit 126 acquires a reduction ratio corresponding to the character size designated by the user from the table.

  For example, consider a case where it is desired to extract characters of 20 points or more from an image. In the image, it is assumed that characters having a size smaller than 20 points such as 10 points and 16 points are included in addition to characters of 20 points or more. If the input image has sufficient resolution for extracting 10-point or 16-point characters, extracting the character area using the input image as it is is more than the unnecessary 20 points from the input image. Even small characters will be extracted. Therefore, the image reduction means 126 is used to reduce the resolution of the image to be processed so that characters with a size smaller than necessary are not extracted.

  Edge detection performed by the edge detection means 121, extraction of character areas performed by the area extraction / separation means 122, and integration of character areas performed by the character area integration means 125 (integration of inverted characters and non-inverted characters) are performed in the first embodiment. It is the same as the form. However, in the second embodiment, edge detection, character region extraction, and character region integration are performed not on the image (original image) input by the image input device 110 but on the image reduced by the image reduction unit 126. . The character area integration unit 125 outputs the character area extracted from the output / reduced image.

  The area matching unit 127 performs matching between the character area extracted from the reduced image and the original image, and extracts a detailed character area from the original image. The area matching unit 127 is configured such that the character area extracted from the reduced image is a character area (non-inverted character) extracted from the first image, or a character area (inverted character) extracted from the second image. Depending on whether or not there is, a pixel having a low feature amount or a pixel having a high feature amount is extracted as a character region from the region of the original image corresponding to the pixel extracted as the character region in the reduced image. Hereinafter, the luminance component of the pixel value is considered as the feature amount. The feature amount may be another component such as a hue.

  FIG. 13 shows an operation procedure. The image input device 110 inputs an image (step S1). The image reducing unit 126 reduces the input image (step S7). The image reduction unit 126 refers to the parameter storage unit 131 and determines an image reduction rate so that the reduced image has a resolution corresponding to the character size of the processing target specified by the user. Alternatively, the reduction ratio may be arbitrarily set by the user.

  The operations from step S2A, S2B to step S6 are the same as the operations in the first embodiment. That is, the edge detection unit 121 detects the first edge and the second edge from the reduced image (steps S2A and S2B). The local binarization unit 123 generates a first image in the vicinity of the first edge, in which a pixel with high luminance is white and a pixel with low luminance is black in the reduced image (step S3A). Further, the local binarization unit 123 generates a second image in the vicinity of the second edge, in which the low-luminance pixel is white and the high-luminance pixel is black in the reduced image (step S3B). The character region separation unit 124 extracts black pixels surrounded by white pixels as character regions in the first image and the second image, respectively (steps S4A and S4B). The character region integration means 125 compares the areas of the character regions at overlapping positions (step S5) and integrates the character regions (step S6).

  The character area integrated by the character area integration unit 125 is a character area extracted from the reduced image. Since the resolution of the reduced image is lower than the resolution of the original image, the extracted character area becomes a roughly extracted character area. The region matching unit 127 performs matching between the character region integrated by the character region integration unit 125 and the image (original image) input in step S1, and extracts a detailed character region at the resolution of the original image (step S1). S8).

  The area matching unit 127 outputs the position of the pixel constituting the character area in the reduced image from the character area integration unit 125 and the character area extracted from the second image or the character area extracted from the first image. Receive information indicating the area. For the character region extracted from the first image, the region matching unit 127 extracts a pixel having low luminance as a character region from the corresponding original image region. For the character region extracted from the second image, the region matching unit 127 extracts pixels having high luminance as the character region from the corresponding original image region.

  FIG. 14A shows a character area extracted from the reduced image, and FIG. 14B shows a corresponding area of the original image. In FIG. 14A, a pixel represented by a dark color corresponds to a pixel extracted as a character area. The numerical value described in the pixel extracted as the character area represents the feature amount (luminance value) of the pixel of the reduced image. The image reduction ratio is 1/3. One pixel of the reduced image corresponds to a 3 × 3 area in the original image. The 2 × 2 area of the reduced image shown in FIG. 14A corresponds to the 6 × 6 area of the original image shown in FIG. In the original image, a pixel indicated by A is a pixel having a luminance value of 14, and a pixel indicated by B is a pixel having a luminance value of 100.

  For example, consider the upper right pixel (luminance value 57) of the four pixels shown in FIG. The area (3 × 3 area) of the original image corresponding to this pixel has four pixels with a luminance value of 14 (the pixel indicated by pixel A in FIG. 14B), and a pixel with a luminance value of 100 (FIG. 14 (FIG. 14 (b)). There are five pixels b). It is assumed that the extracted character region is a character region extracted from the first image, that is, a character region of non-inverted characters in which a pixel having a high luminance and a pixel having a low luminance constitute a character.

  The region matching unit 127 compares the luminance value (luminance value 57) of the pixel extracted as a character region from the reduced image with the luminance value of each pixel in the corresponding region of the original image. The area matching unit 127 determines that a pixel having a luminance value lower than 57 in the corresponding area of the original image is a pixel having a low luminance, and extracts the pixel as a character area. In FIG. 14B, the area matching unit 127 extracts the pixel A as a character area. The area matching unit 127 performs the same processing for other areas. The region matching unit 127 finally extracts, as a character region, pixels indicated by a dark color in FIG. 14B from the original image.

  If the extracted character region is a character region extracted from the second image, that is, a character region of an inverted character in which a pixel with high luminance and a pixel with high luminance constitute a character, the region matching unit 127 The character area is extracted by the reverse operation to that described above. That is, the area matching unit 127 extracts, as a character area, a pixel whose luminance is higher than the luminance value of the pixel extracted as a character area from the reduced image in the corresponding area of the original image.

  In the present embodiment, the image reducing unit 126 reduces the input image. The area matching unit 127 performs matching between the character area extracted from the reduced image and the original image before reduction, and extracts the character area from the original image. When the original image has an excessive resolution with respect to the character size to be processed, a character region of a character having a small size that does not need to be extracted is also extracted. As the resolution is higher, the number of pixels to be processed increases, so that the amount of memory required for processing increases and the time required for extracting the character region also increases. In the present embodiment, edge detection and character area extraction are performed on a reduced image, so that the amount of memory to be used can be reduced compared to the case of using an original image with a high resolution, and processing Speed can be improved. Other effects are the same as those of the first embodiment.

  In each of the above embodiments, the edge detection unit 121 detects the first edge and the second edge according to the direction of change of the feature value, but either edge may be detected. This is because the first edge and the second edge are adjacent to each other, and therefore, by extracting a pixel having a high feature amount surrounded by pixels having a low feature amount in the vicinity of the first edge, the character of the inverted character is also obtained. This is because the region can be extracted, and the character region of the non-inverted character can also be extracted by extracting the pixel with the low feature amount surrounded by the pixel with the high feature amount in the vicinity of the second edge. is there. However, it is considered that a more accurate character region can be extracted by detecting the first edge and the second edge according to the direction of change of the feature amount. The reason for this is that a pixel with a low feature value at the boundary between a pixel with a high feature value and a low pixel becomes the first edge, and a pixel with a high feature value at the boundary between a pixel with a high feature value and a low pixel is the second edge. Therefore, the first edge overlaps with the position of the pixel constituting the non-inverted character, and the second edge overlaps with the position of the pixel constituting the inverted character.

  In each of the above embodiments, the first image and the second image are generated, but either one of the generated binary images may be used. The non-inverted character region and the inverted character region are extracted from pixels having a low feature amount surrounded by pixels having a high feature amount and pixels having a high feature amount surrounded by pixels having a low feature amount. Can be extracted. Therefore, without generating the second image, the first image is used to extract the black pixels surrounded by the white pixels as the character region of the non-inverted characters, and the white pixels surrounded by the black pixels are converted to the inverted characters. May be extracted as a character area. Conversely, without generating the first image, the second image is used to extract white pixels surrounded by black pixels as the character area of non-inverted characters, and the black pixels surrounded by white pixels are inverted. You may extract as a character area of a character.

  An image read by the image input device 110 may be subjected to some image processing and input to the data processing unit 120. For example, an area determination process may be performed before the edge detection unit 121 performs edge detection, and an area where it is clear that there is no character, such as a picture or a photograph, may be deleted from the input image. Further, in the second embodiment, a configuration in which matching between the character region extracted from the reduced image and the original image is not possible is also possible. In other words, if the reduced image has sufficient resolution and character recognition or the like is possible using the character area extracted from the reduced image, the character area roughly extracted from the image output device 140 without matching is used. It may be output.

  Although the present invention has been described based on the preferred embodiment, the image processing apparatus, method, and program of the present invention are not limited to the above embodiment, and various configurations are possible from the configuration of the above embodiment. Those modified and changed as described above are also included in the scope of the present invention.

  INDUSTRIAL APPLICABILITY The present invention can be applied to uses such as a character area extraction device that extracts a character area from a document image with various character decorations, and a program for realizing the character area extraction device in a computer. Further, the present invention can be applied to extraction of a character region in an optical character reading device or a file generation device that outputs a character recognition result to a file.

10: Image processing device 11: Edge detection unit 12: Region extraction / separation unit 100: Image processing device 110: Image input device 120: Data processing unit 121: Edge detection unit 122: Region extraction / separation unit 123: Local binarization unit 124 : Character region separation unit 125: Character region integration unit 126: Image reduction unit 127: Region matching unit 130: Data storage unit 131: Parameter storage unit 140: Image output device

Claims (24)

An edge detection means for detecting an edge based on a change in a feature amount between adjacent pixels from an input image;
A pixel in the vicinity of the pixel of the input image corresponding to the detected edge is divided into a pixel having a high feature value and a pixel having a low feature value based on the feature value, and is surrounded by pixels having a high feature value. An image processing apparatus comprising: a pixel having a low feature amount; and a region extraction / separation unit that extracts a pixel having a high feature amount surrounded by pixels having a low feature amount as a character region. The region extraction / separation means includes
Of the pixels in the vicinity of the pixel of the input image corresponding to the edge, a first image in which the pixel having a low feature amount is black and the pixel having a high feature amount is white, and a pixel having a high feature amount A local binarization unit that generates a second image in which black is used and a pixel having a low feature amount is white, and black pixels surrounded by white pixels are extracted from the first image and the second image. The image processing apparatus according to claim 1, further comprising a character area separating unit that outputs the character area.   When the character area extracted from the first image and the character area extracted from the second image overlap each other, the character that selects one of the overlapping character areas as the character area The image processing apparatus according to claim 2, further comprising region integration means.   The character region integration unit performs labeling on the character region extracted from the first image and the character region extracted from the second image, and a label area for a label at an overlapping position. The image processing apparatus according to claim 3, wherein a character area having a larger is selected as a character area.   The image processing apparatus according to claim 1, further comprising an image reduction unit that reduces an image size of the original image and inputs the reduced image to the edge detection unit.   The image processing apparatus according to claim 5, further comprising a region matching unit that performs matching between the character region extracted from the reduced image and the original image, and extracts a detailed character region from the original image.   The area matching unit is configured to extract a pixel extracted as a character area in the reduced image when a pixel extracted as a character area in the reduced image corresponds to a character area surrounded by pixels having a high feature quantity and having a low feature quantity. Among the pixels of the original image region corresponding to the above, the pixel having a feature amount lower than the feature amount of the pixel of the reduced image is extracted as the detailed character region, and the pixel extracted as the character region in the reduced image is, When corresponding to a character region having a high feature amount surrounded by pixels having a low feature amount, out of the pixels in the original image region corresponding to the pixels extracted as the character region in the reduced image, The image processing apparatus according to claim 6, wherein a pixel having a feature amount higher than a feature amount is extracted as the detailed character region.   The edge detection means detects an edge in the direction in which the feature amount increases as a first edge, detects an edge in a direction in which the feature amount decreases as a second edge, and the region extraction separation means Among the pixels of the input image near the first edge, pixels having a low feature amount surrounded by pixels having a high feature amount are extracted as a character region, and pixels of the input image near the second edge The image processing apparatus according to claim 1, wherein a pixel having a high feature amount surrounded by pixels having a low feature amount is extracted as a character region.   The edge detection means detects, as the first edge, a pixel in which a change in luminance value between adjacent pixels is greater than or equal to a threshold value and changes from a bright luminance value to a dark luminance value from the input image, The image processing apparatus according to claim 8, wherein a pixel in which the change in the luminance value is equal to or greater than a threshold and the luminance value changes from a dark luminance value to a bright luminance value is detected as the second edge.   The edge detection unit detects, as the first edge, a pixel in which a change in hue of an adjacent pixel is greater than or equal to a threshold value and a saturation value is changed from a large value to a small value from the input image. The image processing apparatus according to claim 8, wherein a pixel whose change is greater than or equal to a threshold and whose saturation changes from a small value to a large value is detected as the second edge. A computer detecting an edge from an input image based on a change in a feature amount between adjacent pixels;
The computer classifying a pixel in the vicinity of the pixel of the input image corresponding to the detected edge into a pixel having a high feature value and a pixel having a low feature value based on the feature value;
And a step of extracting, as a character region, a pixel having a low feature amount surrounded by pixels having a high feature amount and a pixel having a high feature amount surrounded by pixels having a low feature amount.   In the step of dividing the pixel into a pixel having a high feature amount and a pixel having a low feature amount, the computer sets the pixel having the low feature amount as black among the pixels in the vicinity of the pixel of the input image corresponding to the edge, and the feature amount is In the step of generating a first image in which a high pixel is white and a second image in which a pixel having a high feature amount is black and a pixel having a low feature amount is white, and extracting the character region, The image processing method according to claim 11, wherein black pixels surrounded by white pixels are extracted as a character region in the first image and the second image.   When the computer is at a position where the character area extracted from the first image and the character area extracted from the second image overlap, any one of the character areas at the overlapping position is defined as a character area. The image processing method according to claim 12, further comprising the step of selecting as:   In the step of selecting the character area, the computer labels the character area extracted from the first image and the character area extracted from the second image, and labels at overlapping positions. The image processing method according to claim 13, wherein a character region having a larger label area is selected as a character region.   The computer according to claim 11, further comprising a step of reducing the image size of the original image and generating a reduced image to be subjected to edge detection in the edge detection step prior to the step of detecting the edge. The image processing method according to any one of the above.   The image processing method according to claim 15, further comprising: a step of matching the character area extracted from the reduced image with the original image and extracting a detailed character area from the original image.   In the step of detecting the edge, the computer detects an edge in a direction in which the feature amount increases as a first edge, and detects an edge in a direction in which the feature amount decreases as a second edge, In the step of classifying the pixel having a high feature amount and the pixel having a low feature amount, the pixel in the vicinity of the pixel of the input image corresponding to the first edge is divided into a pixel having a high feature amount and a pixel having a low feature amount, In the step of dividing the pixels in the vicinity of the pixels of the input image corresponding to the second edge into pixels having high feature values and pixels having low feature amounts, and extracting a character region, among the pixels in the vicinity of the first edge A pixel surrounded by pixels having a high feature value and having a low feature value is extracted as a character area, and a pixel having a high feature value surrounded by pixels having a low feature value among the pixels near the second edge. Extracting as a character area, an image processing method according to any one of claims 11 to 16. On the computer,
From the input image, processing for detecting an edge based on a change in feature amount between adjacent pixels;
Processing to classify a pixel in the vicinity of the pixel of the input image corresponding to the detected edge into a pixel having a high feature value and a pixel having a low feature value based on the feature value;
A program for executing, as a character region, a pixel having a low feature amount surrounded by pixels having a high feature amount and a pixel having a high feature amount surrounded by pixels having a low feature amount.   In the process of discriminating between the high feature amount pixel and the low feature amount pixel, among the pixels in the vicinity of the pixel of the input image corresponding to the edge, the low feature amount pixel is black and the high feature amount pixel is white. In the process of generating the first image and the second image in which the pixel having the high feature amount is black and the pixel having the low feature amount is white and extracting the character region, the first image 19. The program according to claim 18, wherein black pixels surrounded by white pixels are extracted as a character area in the second image.   When the character area extracted from the first image and the character area extracted from the second image are at a position where the computer overlaps, either one of the character areas at the overlapping position is a character area. The program according to claim 19, further causing a process to be selected to be executed.   In the process of selecting the character region, labeling is performed on the character region extracted from the first image and the character region extracted from the second image. The program according to claim 20, wherein a character area having a larger label area is selected as a character area.   19. The computer further includes a step of reducing the image size of the original image and generating a reduced image to be subjected to edge detection in the processing of detecting an edge prior to the step of detecting the edge. The program as described in any one of thru | or 21.   The program according to claim 22, further causing the computer to perform a process of matching a character area extracted from the reduced image with the original image and extracting a detailed character area from the original image.   In the processing for detecting the edge, the edge in the direction in which the feature amount increases is detected as the first edge, and the edge in the direction in which the feature amount decreases is detected as the second edge, and the feature amount is high. In the process of dividing the pixel into the low pixel, the pixel in the vicinity of the pixel of the input image corresponding to the first edge is divided into a pixel having a high feature amount and a pixel having a low feature amount, and the second In the process of dividing the pixels in the vicinity of the pixel of the input image corresponding to the edge into pixels having a high feature amount and pixels having a low feature amount, and extracting the character region, the feature amount among the pixels in the vicinity of the first edge A pixel with a low feature amount surrounded by pixels with a high feature amount is extracted as a character region, and a pixel with a high feature amount surrounded by pixels with a low feature amount among pixels near the second edge is extracted as a character region. Do Program according to any one of claim 18 to 23.

Images