JP2008011484A - Apparatus and method for extracting character and graphic string, program for executing the method, recording medium with the program stored therein - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To extract a region that is a character and graphic string, from an image appropriately and at a high speed. <P>SOLUTION: A character and graphic string region is output which is obtained by applying to image data input by an image data input means, in turn, a means 12 for applying edge extraction processing and binarization processing, an isolate point removing means 13 for removing isolation points, a straight line removing section 141, a brush processing means 14, a morphology processing means 15, a shape analysis means 16 for analyzing a shape, based on a threshold acquired from a threshold storage section 19, and an overlapping region removing means 17. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  The present invention relates to a technique for extracting a pattern such as a character graphic string or a graphic from image data.

  In recent years, digital cameras have been installed in mobile phones, PCs (Personal Computers), etc., and image data can be easily acquired and stored at any time. As a means for further utilizing this function, it is conceivable to recognize a character graphic (that is, a graphic forming a character) sequence appearing in an image from image data and use it as an index for image retrieval.

  For example, as a means for efficiently detecting a character graphic string, a method is known in which a component unrelated to a character is removed from an image step by step, and the last remaining component is detected as a character component (for example, non-character Patent Document 1). This method proposes an algorithm that detects edge components, removes long straight line components and isolated points, takes a circumscribed rectangle of the remaining components, and detects a character graphic string from an analysis result on the shape. Further, this method is a method mainly for drawings and documents.

As a related technique, a technique of automatically selecting a threshold value and performing binarization is also widely known.
Zhaoyang Lu, “Detection of Text Regions from Digital Engineering Drawings”, IEEE Transactions on Pattern Analysis and Machine IntelliIL. 20, no. 4, p. 431-439. Otsu, Nobuyuki, “Automatic threshold selection method based on discrimination and least square criterion”, IEICE Transactions, 1980, Vol. J63-D, no. 4, pp. 349-356.

  Since the above-described method is intended for drawings and the like, for example, a character graphic string is extracted from a natural image photographed by a digital camera (for example, an image in which a signboard with characters is captured in the background or a video including a telop). It is not an optimal method. In particular, when performing a large amount of image indexing or handling video, a character / graphic string must be extracted as fast as possible, and the above-described method is somewhat complicated and causes a problem in terms of processing speed.

  The present invention has been made based on the above-described problems, and executes a character graphic string extraction device, a character graphic string extraction method, and a method thereof that can appropriately and quickly extract an area that is a character graphic string from a natural image or video. It is to provide a program to be recorded and a recording medium on which the program is recorded.

  In order to solve the above problems, the present invention provides a character / graphic string extraction apparatus for extracting a region for forming a character / graphic string included in image data, wherein the image data is converted into an image. Image data input means for inputting from the data acquisition means, and edge extraction processing is performed on the input image data, and further, first-stage image data is generated by performing binarization processing on the edge extracted image data Edge extracting means for removing the first value pixels isolated from the other first value pixels with respect to the first stage image data, and generating second stage image data, and the second stage image A straight line removing unit that detects a region of first value pixels that are linearly continuous from the data, and further generates third-stage image data in which the first value pixels constituting the region are removed; A brush processing means for reading out a threshold value defining the threshold value from the threshold value storage unit, performing a brush process on the third-stage image data based on the threshold value, and generating fourth-stage image data; and the fourth-stage image data On the other hand, a morphological processing means for generating fifth-stage image data by performing erosion and dilation, and the first-value pixel area is labeled with respect to the fifth-stage image data, and the same area is regarded as one area. First, a first circumscribing rectangle of each region is calculated, and a first determination for determining true when the number of first value pixels in the first circumscribing rectangle exceeds the upper limit of the number of first value pixels, 2nd determination which judges as true when the number of first value pixels in one circumscribed rectangle exceeds the lower limit value of the first value pixels, the short side length of the first circumscribed rectangle exceeds the lower limit value of the short side length A third determination that determines that the case is true, the third determination Fourth determination for determining true when the ratio between the number of first value pixels and the number of second value pixels in the circumscribed rectangle exceeds the lower limit value of the ratio, and the inside of the first circumscribed rectangle in the input image data The first circumscribed rectangle determined to be true in any one of the fifth determinations in which the contrast of the portion corresponding to 1 exceeds the lower limit of the contrast is determined to be true, and is calculated as the second circumscribed rectangle. The second circumscribed rectangle is removed from the first circumscribed rectangle and the remaining first circumscribed rectangle is generated as the third circumscribed rectangle; and the third circumscribed rectangle in the inclusion relationship, An overlapping area removing unit that removes the contained third circumscribed rectangle and regards the remaining third circumscribed rectangle as a fourth circumscribed rectangle, and regards the fourth circumscribed rectangle as a character graphic string region. A character graphic string region output means And

  The invention according to claim 2 is a character graphic string extraction method for extracting a region for forming a character graphic string included in image data, the image data input step for inputting the image data from an image data acquisition means, An edge extraction step for performing an edge extraction process on the input image data, and further generating a first stage image data obtained by performing a binarization process on the edge extracted image data, and the first stage image data An isolated point removing step for removing the first value pixel isolated from the other first value pixels to generate second stage image data, and a region of the first value pixels linearly continuous from the second stage image data. The threshold value is defined as a straight line removing step for generating third-stage image data from which the first value pixels constituting the region are removed, and a neighborhood defining the pixel of interest. A step of performing a brush process on the third-stage image data based on the threshold value, generating a fourth-stage image data, and performing erosion and dilation on the fourth-stage image data. Then, the morphology processing step for generating the fifth stage image data and the first value pixel area are labeled with respect to the fifth stage image data, and the first area of each area is regarded as one area. A circumstance rectangle is calculated, and a first determination for determining as true when the number of first value pixels in the first circumscribed rectangle exceeds an upper limit value of the first value pixel number, a first value in the first circumscribed rectangle Second determination for determining true when the number of pixels exceeds the lower limit of the first number of pixels, and determining as true when the short side length of the first circumscribed rectangle exceeds the lower limit of the short side length 3 determination, the first in the first circumscribed rectangle Fourth determination for determining true when the ratio between the number of pixels and the number of second-value pixels exceeds the lower limit value of the ratio. In the input image data, the contrast of the portion corresponding to the inside of the first circumscribed rectangle is The first circumscribed rectangle determined as true in any one of the fifth determinations in which the case where the lower limit value of the contrast is exceeded is determined as true is regarded as the second circumscribed rectangle, and is calculated from the calculated first circumscribed rectangle. A shape analysis step for removing the second circumscribed rectangle and generating the remaining first circumscribed rectangle as a third circumscribed rectangle, and a third circumscribed rectangle that is included among the third circumscribed rectangles in an inclusive relationship And an overlapping area removing step for generating the remaining third circumscribed rectangle as a fourth circumscribed rectangle, and outputting a character / graphic string area as if the fourth circumscribed rectangle is regarded as a character / graphic string area And a step.

  The invention described in claim 3 is a character graphic string extraction program, wherein the character graphic string extraction method according to claim 2 is described as a computer program executable by a computer.

  The invention described in claim 4 is a recording medium, wherein the character / graphic string extraction method described in claim 2 is described in a computer-executable program and the program is recorded.

  The invention according to claim 5 is a character / graphic string extraction device for extracting a region forming a character / graphic string included in image data, the image data input means for inputting the image data from an image data acquisition means, Edge extraction means for performing edge extraction processing on the input image data and generating first-stage image data obtained by binarizing the edge-extracted image data; and the first-stage image data An isolated point removing unit that removes first value pixels isolated from other first value pixels to generate second stage image data, and a region of first value pixels that are linearly continuous from the second stage image data And a straight line removing unit that generates third-stage image data from which the first value pixels constituting the area are removed, and a threshold value that defines a neighborhood related to the target pixel is read from the threshold storage unit. The third stage image data is brushed based on the threshold value to generate the fourth stage image data, and the fourth stage image data is subjected to erosion and dilation. , A morphology processing means for generating fifth-stage image data, and a first value pixel area for the fifth-stage image data, labeling the same area as one area, and a first circumscribed rectangle of each area , And a first determination that determines true when the number of first value pixels in the first circumscribed rectangle exceeds the upper limit of the number of first value pixels, the number of first value pixels in the first circumscribed rectangle Is determined to be true when the value exceeds the lower limit of the number of first value pixels, and is determined as true when the short side length of the first circumscribed rectangle exceeds the lower limit of the short side length. , The first value pixel number and the second value pixel in the first circumscribed rectangle A fourth determination for determining the case where the ratio exceeds the lower limit of the ratio as true, and the contrast of the portion corresponding to the inside of the first circumscribed rectangle in the input image data exceeds the lower limit of the contrast The first circumscribed rectangle determined to be true in the determination of any one of the fifth determinations to determine as true is regarded as the second circumscribed rectangle, and the second circumscribed rectangle is removed from the calculated first circumscribed rectangle, Shape analysis means for generating the remaining first circumscribed rectangle as a third circumscribed rectangle, and the third circumscribed rectangle included in the third circumscribed rectangle in the inclusion relation is removed, and the remaining first circumscribed rectangle is removed. Overlapping area removing means for generating three circumscribed rectangles as fourth circumscribed rectangles, ease of binarization of circumscribed rectangular areas of the fourth circumscribed rectangles, little variation in stroke width, stroke Maximum value of width, first value pixel after binarization The complexity of the boundary between the region and the second value pixel region and the maximum stroke length are each digitized, and those determined as circumscribed rectangular regions that do not look like characters are removed from the fourth circumscribed rectangle, and the remaining second A character-likeness analysis unit that generates the four circumscribed rectangles as a fifth circumscribed rectangle; and a character / graphic string region output unit that outputs the fifth circumscribed rectangle as a character / graphic column region. Features.

  The invention according to claim 6 is a character / graphic string extraction method for extracting a region for forming a character / graphic string included in image data, the image data input step for inputting the image data from an image data acquisition means, An edge extraction step for performing an edge extraction process on the input image data, and further generating a first stage image data obtained by performing a binarization process on the edge extracted image data, and the first stage image data An isolated point removing step for removing the first value pixel isolated from the other first value pixels to generate second stage image data, and a region of the first value pixels linearly continuous from the second stage image data. The threshold value is defined as a straight line removing step for generating third-stage image data from which the first value pixels constituting the region are removed, and a neighborhood defining the pixel of interest. A step of performing a brush process on the third-stage image data based on the threshold value, generating a fourth-stage image data, and performing erosion and dilation on the fourth-stage image data. Then, the morphology processing step for generating the fifth stage image data and the first value pixel area are labeled with respect to the fifth stage image data, and the first area of each area is regarded as one area. A circumstance rectangle is calculated, and a first determination for determining as true when the number of first value pixels in the first circumscribed rectangle exceeds an upper limit value of the first value pixel number, a first value in the first circumscribed rectangle Second determination for determining true when the number of pixels exceeds the lower limit of the first number of pixels, and determining as true when the short side length of the first circumscribed rectangle exceeds the lower limit of the short side length 3 determination, the first in the first circumscribed rectangle Fourth determination for determining true when the ratio between the number of pixels and the number of second-value pixels exceeds the lower limit value of the ratio. In the input image data, the contrast of the portion corresponding to the inside of the first circumscribed rectangle is The first circumscribed rectangle determined as true in any one of the fifth determinations in which the case where the lower limit value of the contrast is exceeded is determined as true is regarded as the second circumscribed rectangle, and is calculated from the calculated first circumscribed rectangle. A shape analysis step for removing the second circumscribed rectangle and generating the remaining first circumscribed rectangle as a third circumscribed rectangle, and a third circumscribed rectangle that is included among the third circumscribed rectangles in an inclusive relationship And the overlapping region removal step for generating the remaining third circumscribed rectangle as a fourth circumscribed rectangle, the ease of binarization of the circumscribed rectangular region of the fourth circumscribed rectangle, and the stroke width Less dispersion, maximum stroke width The binarized first-value pixel area and the second-value pixel area are each digitized in terms of the complexity of the boundary and the maximum stroke length. Character likelihood analysis step that is generated by considering the remaining fourth circumscribed rectangle as the fifth circumscribed rectangle, and removing the fourth circumscribed rectangle as the fifth circumscribed rectangle, and the character that is output by regarding the fifth circumscribed rectangle as the character graphic string region And a graphic sequence region output step.

  The invention described in claim 7 is a character graphic string extraction program, wherein the character graphic string extraction method according to claim 6 is described as a computer program executable by a computer.

  The invention described in claim 8 is a recording medium, wherein the character / graphic string extraction method described in claim 6 is described in a computer-executable program and the program is recorded.

  According to the first, second, third, and fourth aspects of the present invention, a simplified shape analysis process is performed on an image subjected to the above-described edge extraction, binarization process, isolated point removal, brush process, and morphology process. Can be executed.

  According to the inventions of the fifth, sixth, seventh, and eighth aspects, a simplified shape analysis process is performed on an image that has been subjected to the above-described edge extraction, binarization process, isolated point removal, brush process, and morphology process. Characteristic analysis processing can be executed.

  As described above, according to the first, second, third, and fourth aspects of the invention, since the shape analysis processing is simplified, the character / graphic string extraction region is appropriately and quickly extracted from the image as the entire character / graphic string extraction processing. it can. In addition, it is possible to suppress erroneous extraction of the non-character graphic string region, and to reduce the burden of subsequent recognition processing.

  According to the fifth, sixth, seventh, and eighth inventions, it is possible to suppress erroneous extraction of non-character string regions and to reduce the burden of subsequent recognition processing, as compared with the first, second, third, and fourth inventions.

  These can contribute to the character recognition technology field.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

  The configuration of the character graphic string extraction apparatus in this embodiment will be described with reference to FIG. The character / graphic string extraction apparatus according to the present embodiment includes an image data input unit 11, an image processing unit 1, a character / graphic string region output unit 18, and a threshold storage unit 19. The image processing unit 1 includes an edge extraction unit 12, an isolated point removal unit 13, a straight line removal unit 141, a brush processing unit 14, a morphology processing unit 15, a shape analysis unit 16, and an overlapping region removal unit 17. . The threshold storage unit 19 may be provided with a means for inputting and storing a threshold value, or a fixed value may be stored in the threshold storage unit 19 in advance. The threshold is a threshold for recognizing a pattern including, for example, a character graphic (or character graphic string), a graphic (for example, a trademark), or the like.

  The image data input unit 11 receives natural image data (that is, multi-valued image data) from an image data acquisition unit (for example, a digital camera or a database storing image data) and transmits the natural image data to the edge extraction unit 12. When natural image data is input from a video, each frame image (that is, image data) in the video is input from the image data input unit 11.

  The edge extraction unit 12 performs edge extraction processing on the transmitted image data, generates image data obtained by performing binarization processing on the extracted image data, and transmits the image data to the isolated point removal unit 13. For the edge extraction process, for example, a Sobel operator, a Laplacian operator, or the like is used. In the binarization process, for example, a method using a fixed threshold value, a method using an adaptive threshold value based on a brightness histogram in a local window, or the like is used. In the following description, an area obtained by binarizing the edge area will be described as a black pixel (that is, a pixel having a pixel value “0”; a first value pixel).

  The isolated point removal unit 13 transmits the image data generated as a result of deleting one isolated point black pixel from the image data after edge extraction to the straight line removal unit 141.

The straight line removal unit 141 detects a place where black pixels continue linearly from the image data after the isolated point black pixels are removed, and removes all black pixels constituting them. Specifically, first, the image is rotated (or skew-transformed) by the rotation angle θ ° in the world coordinate system. It is assumed that the rotated image is also a binary image. Next, the image is scanned (inspected linearly) in the vertical direction (world coordinate system y-axis direction) or the horizontal direction (world coordinate system x-axis direction), and the run length L of the black pixel is measured. A case where the length L is larger than a specific threshold value n ₁ is regarded as a straight line. The threshold value n ₁ is read from the threshold value storage unit 19. Next, the rotation angle θ ° is changed in several steps and processed in the same manner. Then, all the constituent pixels regarded as the straight line are removed. Finally, the image data at the stage where the straight line has been removed is generated and transmitted to the brush processing unit 14.

  The brush processing unit 14 performs a process of extending all the black pixels transmitted from the straight line removal unit 141 in the vertical and horizontal directions (image coordinate system y direction x direction) to a certain distance (number of pixels). That is, the calculation based on the following rules is performed on the entire image data transmitted from the straight line removal unit 141. The brushing process is a process of extending all target pixels radially to a certain distance (number of pixels). Then, the image data as the calculation result is generated and transmitted to the morphology processing unit 15.

First, any one of the pixels within the range of n ₂ pixels above (+ y direction in the image coordinate system) and n ₂ pixels below (−y direction of the image coordinate system) from the target pixel is a black pixel. If so, the pixel of interest is also black. Note that n ₂ is a threshold value that defines the neighborhood related to the target pixel in the brushing process, and indicates a positive value. The threshold value n ₂ is read from the threshold value storage unit 19.

Thereafter, any one of the pixels within the range corresponding to the number of n ₂ pixels on the left (−x direction of the image coordinate system) and the number of n ₂ pixels on the right (the + x direction of the image coordinate system) from the target pixel is a black pixel. If so, the pixel of interest is also black.

  The above is the brushing process in the present embodiment.

  The morphology processing unit 15 performs erosion and dilation on the image data transmitted from the brush processing unit 14 (that is, performs a morphology process), generates the applied image data, and transmits the generated image data to the shape analysis unit 16.

  The shape analysis unit 16 performs the following processing as shape analysis.

  First, black pixel labeling is performed on the image data transmitted from the morphology processing unit 15, and a circumscribed rectangular area (hereinafter simply referred to as a circumscribed rectangle) of each area is calculated (extracted) by regarding the same label area as one area. ). Note that the circumscribed rectangle also allows a rectangle (rectangle) rotated obliquely.

Then, for each circumscribed rectangle, circumscribed rectangles that apply to any one of the following formulas 1 to 5 are removed, and circumscribed rectangles that do not apply at all are extracted. Note that n ₃ is a threshold value that defines the upper limit value of the number of black pixels in the circumscribed rectangle, and is a positive value. n ₄ is a threshold value that defines a lower limit value of the number of black pixels in the circumscribed rectangle, and is a positive value. n ₅ is a threshold value that defines the lower limit value of the short side of the circumscribed rectangle, and is a positive value. n ₆ is a threshold value defining a lower limit value of the ratio of the number of black pixels and the number of white pixels (second value pixels) in the circumscribed rectangle, and is a positive value. n ₇ is a threshold value that defines the lower limit value of the contrast of the original image (that is, natural image data) in the circumscribed rectangle, and is a positive value. The contrast is measured with an original image before binarization in a rectangle, and is defined by, for example, the variance of a brightness histogram. The thresholds n ₃ to n ₇ are read from the threshold storage unit 19 and used.

Here, P _b is the number of black pixels in the rectangle, L _s is the short side of the rectangle, P _w is the number of white pixels in the rectangle, and P _c is the contrast of the original image in the rectangle.

  The extracted circumscribed rectangle is transmitted to the overlapping area removing unit 17.

  The overlapping area removing unit 17 examines the overlap of the extracted circumscribed rectangles, removes the smaller one (ie, the included circumscribed rectangle) for those that completely include, and converts the remaining circumscribed rectangle into a character figure. The data is transmitted to the column area output unit 18.

  The character graphic string area output unit 18 outputs information about the remaining circumscribed rectangle (for example, the length of the long side of the rectangle, the length of the short side, and the rotation angle) as the character graphic area. For example, the circumscribed rectangle is superimposed on the original image (that is, the image input by the image input unit 11) and displayed on the display device.

  A character graphic string extraction method according to the present embodiment will be described with reference to FIG. In the following description, the same reference numerals as those in FIG. 1 are omitted.

  First, natural image data is input from the image data acquisition means (S101).

  Next, edge extraction processing is performed on the input natural image data, and further binarization processing is performed on the edge-extracted image data (S102).

  Next, one isolated point black pixel is deleted from the image data subjected to the edge extraction process and the binarization process (S103).

  Next, the above-described straight line removal is performed on the image data from which the isolated point black pixels have been deleted (S1041).

  Next, the above-described brushing process is performed on the entire image data from which straight lines have been removed (S104).

  Next, the above-described morphology process is performed on the image data subjected to the brush process (S105).

  Next, the above-described shape analysis is performed on the image data that has been subjected to morphology processing, and a circumscribed rectangular region is extracted (S106).

  Then, the overlap of the extracted circumscribed rectangular areas is examined, and those that are completely included are removed (S107). That is, the remaining circumscribed rectangular area is an area that matches a pattern (for example, a character graphic string pattern) based on the threshold value stored in the threshold value storage unit 19.

  Note that a storage medium that records a program code of software that realizes the functions of the present embodiment is supplied to a system or apparatus, and a CPU (Central Processing Unit) (or MPU (Microprocessing Unit)) of the system or apparatus stores the storage medium. This can also be realized by reading out and executing the program code stored in. In that case, the program code itself read from the storage medium realizes the functions of the above-described embodiment, and a storage medium storing the program code, for example, a CD-ROM (Compact Disk Read Only Memory), DVD-ROM (Digital Versatile Disk Read Only Memory), CD-R (Compact Disk Recordable), CD-RW (Compact Disk ReWriteable), MO (Magneto-Optical Disk), MO (Magneto-Optical Disk) To do.

  In addition, in the case where the accuracy of character / graphic detection is more important than the speedup, an embodiment in which a “characteristic analysis unit” is added after the overlapping region removal unit (for example, the configuration of FIGS. 3 and 4) is also conceivable.

  3, the image data input unit 21 to the overlap region removal unit 27, the straight line removal unit 241 and the character / graphic string region output unit 29 are the image data input unit 11 to the overlap region removal unit 17, the straight line removal unit 141 and the character / figure graphic. Each of the column area output units 18 has the same function.

  The character likelihood analysis unit 28 calculates five types of character likelihood reference values for each of the extracted character graphic region candidates (for example, circumscribed rectangles left by the overlapping region removal unit 27), and determines that the character graphic region is not a character graphic region. Are excluded. Here, all the character / graphic region candidates are detected in a rectangular shape, and the size is normalized in advance so that the short side of the rectangle becomes S. Note that the value of S is determined in advance.

  The first character-likeness reference value is obtained by quantifying the ease of binarization.

  First, a brightness histogram as shown in FIG. 5 is calculated for one character graphic region candidate. In FIG. 5, the x-axis is lightness and the y-axis is frequency.

  Next, a threshold value for binarization is obtained from the histogram. Examples of the threshold calculation method include the threshold calculation method of Non-Patent Document 2.

  Next, two peak positions (for example, the left peak position Pl and the right peak position Pr in FIG. 5) are detected from the histogram. For example, it can be calculated as a position indicating the maximum frequency in each of two regions separated by a threshold (for example, threshold th in FIG. 5).

  Then, the frequency in the range of the width w1 is obtained as a cumulative value before and after each peak position, and the percentage of the whole is examined. Note that the value of the width w1 is determined in advance.

  Here, the ratio calculated from the periphery of the left peak position is referred to as PerL, and the ratio calculated from the right peak is referred to as PerR. If PerL and PerR are smaller than the predetermined Thre1 (PerL <Thre1 and PerR <Thre1), it is regarded as not a character graphic area and is excluded from the candidates.

  The second character-likeness reference value is obtained by quantifying the small dispersion of the stroke width. Note that a stroke refers to a line that constitutes a character.

  First, a binarized character / graphic area candidate is generated by binarizing the area for one character / graphic area candidate.

  Next, the following processing is performed on one black pixel.

  As a process A1, black pixel scanning is performed in a vertical direction, a horizontal direction, a lower left-upper right direction, and an upper left-lower right direction with respect to a certain black pixel, and run lengths in four directions are measured.

  As the process A2, the numerical value of the run length indicating the smallest length is defined as “stroke width” and voted for the “stroke width histogram” as shown in FIG. In FIG. 6, the x-axis is the stroke width and the y-axis is the frequency.

  Processing A1 and processing A2 are performed on all black pixels.

  Next, the peak position P (for example, the peak position P in FIG. 6) is obtained from the stroke width histogram. The peak position P is obtained by searching for a position with the highest frequency.

  Next, at the peak position, the frequency in the range of the width w2 before and after is obtained as a cumulative value, and what percentage is occupied (ratio; PerW1) is examined. Note that the value of the width w2 is determined in advance. PerW2 is obtained by applying the above-described processing for obtaining PerW1 to a binarized character graphic region candidate obtained by reversing black and white. PerW1 and PerW2 are compared, and the larger one is adopted as PerW. Note that the correspondence between the white pixel or black pixel, the character component, and the background component at this time is also used below.

  Finally, if the obtained PerW is smaller than the predetermined Thre2 (PerW <Thre2), it is regarded as not a character graphic area and is excluded from the candidates.

  The third character-likeness reference value is a numerical value of the maximum stroke width. All the above stroke widths are calculated with black pixels (black pixels which are character components), and their maximum values are obtained. The maximum value is obtained as a ratio to the short side length S of the rectangular area as a ratio, and is defined as PerD. If PerD is larger than the predetermined Thre3 (PerD> Thre3), it is regarded as not a character graphic area and is excluded from the candidates.

  The fourth character-likeness reference value is obtained by quantifying the complexity of the boundary between the binarized black pixel (character pixel black pixel) region and the white pixel (background component white pixel) region. is there.

  First, the area is binarized for one character / graphic area candidate.

  Next, pixels corresponding to the boundary between the black pixel region and the white pixel region are detected, and the number thereof is totalized. PerF is determined as a percentage of the total number of pixels. If PerF is larger than the predetermined Thre4 (PerF> Thre4), it is excluded from the candidates as not being a character graphic area.

  The fifth character-likeness reference value is a numerical value of the maximum stroke length.

  First, the area is binarized for one character / graphic area candidate.

  Next, the following processing is performed on one black pixel.

  As a process B1, a black pixel is scanned in a vertical direction, a horizontal direction, a lower left-upper right direction, and an upper left-lower right method with respect to a certain black pixel (character pixel black pixel). Measure each.

  As the process B2, the numerical value of the run length indicating the largest length is defined as the “stroke length”.

  Processing B1 and processing B2 are performed for all black pixels, and the maximum value is obtained. The maximum value is obtained as a ratio with respect to the short side length S of the rectangular region, and is defined as PerS. If PerS is larger than the predetermined Thre5 (PerS> Thre5), it is regarded as not a character graphic area and is excluded from the candidates.

  Information relating to the circumscribed rectangle left by the combination of the character quality reference values described above is output by the character graphic string region output unit 29.

  The processing of this embodiment will be described with reference to FIG. The image data input step S201 to the overlapping area removal step S207 and the straight line removal step S2041 in FIG. 4 perform the same processes as the image data input step S101 to the overlapping area removal step S107 and the straight line removal step S1041, respectively.

  Character character analysis step S208 in FIG. 4 is a step of performing processing similar to that of the character character analysis unit 28 described above. That is, in each of the extracted character / graphic region candidates, the above-described five types of character-likeness reference values are calculated, and those determined not to be character / character regions are excluded.

  The S value, the Thre1 to Thre5 values, and the w1 and w2 values are read from the threshold storage unit 19 and used.

  As described above, the present embodiment is a method for extracting a character graphic string region from image data, and performs processing in the order of edge extraction, isolated point removal, straight line removal, brush processing, morphology processing, shape analysis, and overlapping region removal. I do.

  Further, according to the present embodiment, the shape analysis (S106) of the partial area extracted as the character component can be processed into the following five simplified processes.

  In the first process, an area that is larger than a specific threshold is deleted.

  In the second process, an area that is smaller than a specific threshold is deleted.

  The third process deletes an area that is too narrow than a specific threshold.

  In the fourth process, an area in which the value in the circumscribed rectangle (the value obtained by dividing the number of black pixels by the number of white pixels) is too small than a specific threshold is deleted.

  The fifth process deletes an area where the contrast is extremely smaller than a specific threshold value.

  Since both of these processes are simple processes, a character / graphic string can be extracted from a natural image at a very high speed.

For example, if a character code (for example, JIS (Japan Industrial Standard) code) is assigned to the threshold set of the above-described threshold values n ₃ to n ₇ , it can be handled as a character string (for example, text) handled by a computer.

  Although the present invention has been described in detail only for the specific examples described above, it is obvious to those skilled in the art that various changes and modifications are possible within the scope of the technical idea of the present invention. Such variations and modifications are naturally within the scope of the claims.

For example, as a modification of the present embodiment, a threshold value storage unit that manages a plurality of threshold value sets of threshold values n ₁ to n ₇ may be used. That is, one threshold set is previously selected from the plurality of threshold sets, and each threshold is read from the threshold set.

The lineblock diagram of the character graphic sequence extraction device in this embodiment. The flowchart which shows the character figure string extraction method in this embodiment. The block diagram of the character figure string extraction apparatus in a present Example. The flowchart which shows the character figure string extraction method in a present Example. The figure which shows an example of the brightness histogram in a present Example. The figure which shows an example of the stroke width histogram in a present Example.

Explanation of symbols

DESCRIPTION OF SYMBOLS 1,1 '... Image processing part 11, 21 ... Image data input part 12, 22 ... Edge extraction part 13, 23 ... Isolated point removal part 14, 24 ... Brush processing part 15, 25 ... Morphology processing part 16, 26 ... Shape Analyzing unit 17, 27 ... Overlapping region removing unit 18, 29 ... Character graphic string region output unit 19 ... Threshold storage unit 28 ... Characteristicness analyzing unit 141, 241 ... Straight line removing unit P ... Peak position Pl ... Left peak position Pr ... Right Peak position th ... Threshold

Claims (8)

A character graphic string extraction device for extracting a region forming a character graphic string included in image data,
Image data input means for inputting the image data from image data acquisition means;
An edge extraction unit that performs an edge extraction process on the input image data, and further generates a first stage image data obtained by performing a binarization process on the edge extracted image data;
Isolated point removing means for removing first value pixels isolated from other first value pixels with respect to the first step image data to generate second step image data;
Straight line removing means for detecting a region of first value pixels that are linearly continuous from the second step image data, and further generating third step image data from which the first value pixels constituting the region are removed;
A brush processing means for reading out a threshold value defining a neighborhood related to the target pixel from the threshold value storage unit, performing a brush process on the third-stage image data based on the threshold value, and generating a fourth-stage image data;
Morphology processing means for performing erosion and dilation on the fourth stage image data to generate fifth stage image data;
For the fifth stage image data, the first value pixel area is labeled, the same area is regarded as one area, and the first circumscribed rectangle of each area is calculated,
A first determination for determining true when the number of first value pixels in the first circumscribed rectangle exceeds the upper limit of the number of first value pixels;
A second determination for determining as true when the number of first value pixels in the first circumscribed rectangle exceeds a lower limit of the number of first value pixels;
A third determination for determining true when the short side length of the first circumscribed rectangle exceeds the lower limit of the short side length;
A fourth determination for determining as true when the ratio between the number of first value pixels and the number of second value pixels in the first circumscribed rectangle exceeds a lower limit value of the ratio;
A fifth determination for determining as true when the contrast of the portion corresponding to the inside of the first circumscribed rectangle in the input image data exceeds a lower limit value of contrast;
The first circumscribed rectangle determined to be true in any one of the above is regarded as the second circumscribed rectangle, the second circumscribed rectangle is removed from the calculated first circumscribed rectangle, and the remaining first circumscribed rectangle is replaced with the third circumscribed rectangle. A shape analysis means that is generated by considering it as a circumscribed rectangle;
An overlapping area removing unit that generates the third circumscribed rectangle of the third circumscribed rectangle in the inclusion relation by removing the included third circumscribed rectangle and considering the remaining third circumscribed rectangle as the fourth circumscribed rectangle;
A character / figure string region output means for outputting the fourth circumscribed rectangle as a character / figure string region;
A character graphic string extraction device comprising: A character graphic string extraction method for extracting a region forming a character graphic string included in image data,
An image data input step of inputting the image data from an image data acquisition means;
An edge extraction step of performing an edge extraction process on the input image data, and further generating a first stage image data obtained by performing a binarization process on the image data extracted from the edge;
An isolated point removing step of removing first value pixels isolated from other first value pixels with respect to the first step image data, and generating second step image data;
A straight line removal step of detecting a region of first value pixels linearly continuous from the second step image data, and further generating third step image data from which the first value pixels constituting the region are removed;
A brush processing step of reading out a threshold value defining a neighborhood related to the target pixel from the threshold value storage unit, performing a brush process on the third-stage image data based on the threshold value, and generating a fourth-stage image data;
Morphology processing step for generating fifth stage image data by performing erosion and dilation on the fourth stage image data;
For the fifth stage image data, the first value pixel area is labeled, the same area is regarded as one area, and the first circumscribed rectangle of each area is calculated,
A first determination for determining true when the number of first value pixels in the first circumscribed rectangle exceeds the upper limit of the number of first value pixels;
A second determination for determining as true when the number of first value pixels in the first circumscribed rectangle exceeds a lower limit of the number of first value pixels;
A third determination for determining true when the short side length of the first circumscribed rectangle exceeds the lower limit of the short side length;
A fourth determination for determining as true when the ratio between the number of first value pixels and the number of second value pixels in the first circumscribed rectangle exceeds a lower limit value of the ratio;
A fifth determination for determining as true when the contrast of the portion corresponding to the inside of the first circumscribed rectangle in the input image data exceeds a lower limit value of contrast;
The first circumscribed rectangle determined to be true in any one of the above is regarded as the second circumscribed rectangle, the second circumscribed rectangle is removed from the calculated first circumscribed rectangle, and the remaining first circumscribed rectangle is replaced with the third circumscribed rectangle. A shape analysis step that is generated by considering it as a circumscribed rectangle;
An overlapping region removing step of generating the third circumscribed rectangle of the third circumscribed rectangle in the inclusion relation by removing the included third circumscribed rectangle and considering the remaining third circumscribed rectangle as the fourth circumscribed rectangle;
A character graphic string region output step for outputting the fourth circumscribed rectangle as a character graphic string region;
A character graphic string extraction method characterized by comprising:   A character graphic string extraction program characterized in that the character graphic string extraction method according to claim 2 is described as a computer program executable by a computer.   A recording medium comprising the character graphic string extraction method according to claim 2 described in a computer-executable program and the program recorded therein. A character graphic string extraction device for extracting a region forming a character graphic string included in image data,
Image data input means for inputting the image data from image data acquisition means;
An edge extraction unit that performs an edge extraction process on the input image data, and further generates a first stage image data obtained by performing a binarization process on the edge extracted image data;
Isolated point removing means for removing first value pixels isolated from other first value pixels with respect to the first step image data to generate second step image data;
Straight line removing means for detecting a region of first value pixels that are linearly continuous from the second step image data, and further generating third step image data from which the first value pixels constituting the region are removed;
A brush processing means for reading out a threshold value defining a neighborhood related to the target pixel from the threshold value storage unit, performing a brush process on the third-stage image data based on the threshold value, and generating a fourth-stage image data;
Morphology processing means for performing erosion and dilation on the fourth stage image data to generate fifth stage image data;
For the fifth stage image data, the first value pixel area is labeled, the same area is regarded as one area, and the first circumscribed rectangle of each area is calculated,
A first determination for determining true when the number of first value pixels in the first circumscribed rectangle exceeds the upper limit of the number of first value pixels;
A second determination for determining as true when the number of first value pixels in the first circumscribed rectangle exceeds a lower limit of the number of first value pixels;
A third determination for determining true when the short side length of the first circumscribed rectangle exceeds the lower limit of the short side length;
A fourth determination for determining as true when the ratio between the number of first value pixels and the number of second value pixels in the first circumscribed rectangle exceeds a lower limit value of the ratio;
A fifth determination for determining as true when the contrast of the portion corresponding to the inside of the first circumscribed rectangle in the input image data exceeds a lower limit value of contrast;
The first circumscribed rectangle determined to be true in any one of the above is regarded as the second circumscribed rectangle, the second circumscribed rectangle is removed from the calculated first circumscribed rectangle, and the remaining first circumscribed rectangle is replaced with the third circumscribed rectangle. A shape analysis means that is generated by considering it as a circumscribed rectangle;
An overlapping area removing unit that generates the third circumscribed rectangle of the third circumscribed rectangle in the inclusion relation by removing the included third circumscribed rectangle and considering the remaining third circumscribed rectangle as the fourth circumscribed rectangle;
Of the fourth circumscribed rectangle, the ease of binarization of the circumscribed rectangular area, the small dispersion of the stroke width, the maximum stroke width, the first value pixel area and the second value after binarization The complexity of the boundary with the pixel area and the maximum stroke length are each digitized, and those that are determined to be non-character circumscribed rectangle areas are removed from the fourth circumscribed rectangle, and the remaining fourth circumscribed rectangle is replaced with the fourth circumscribed rectangle. 5 character character analysis means to be generated by considering it as a circumscribed rectangle;
A character / figure string region output means for outputting the fifth circumscribed rectangle as a character / figure string region;
A character graphic string extraction device comprising: A character graphic string extraction method for extracting a region forming a character graphic string included in image data,
An image data input step of inputting the image data from an image data acquisition means;
An edge extraction step of performing an edge extraction process on the input image data, and further generating a first stage image data obtained by performing a binarization process on the image data extracted from the edge;
An isolated point removing step of removing first value pixels isolated from other first value pixels with respect to the first step image data, and generating second step image data;
A straight line removal step of detecting a region of first value pixels linearly continuous from the second step image data, and further generating third step image data from which the first value pixels constituting the region are removed;
A brush processing step of reading out a threshold value defining a neighborhood related to the target pixel from the threshold value storage unit, performing a brush process on the third-stage image data based on the threshold value, and generating a fourth-stage image data;
Morphology processing step for generating fifth stage image data by performing erosion and dilation on the fourth stage image data;
For the fifth stage image data, the first value pixel area is labeled, the same area is regarded as one area, and the first circumscribed rectangle of each area is calculated,
A first determination for determining true when the number of first value pixels in the first circumscribed rectangle exceeds the upper limit of the number of first value pixels;
A second determination for determining as true when the number of first value pixels in the first circumscribed rectangle exceeds a lower limit of the number of first value pixels;
A third determination for determining true when the short side length of the first circumscribed rectangle exceeds the lower limit of the short side length;
A fourth determination for determining as true when the ratio between the number of first value pixels and the number of second value pixels in the first circumscribed rectangle exceeds a lower limit value of the ratio;
A fifth determination for determining as true when the contrast of the portion corresponding to the inside of the first circumscribed rectangle in the input image data exceeds a lower limit value of contrast;
The first circumscribed rectangle determined to be true in any one of the above is regarded as the second circumscribed rectangle, the second circumscribed rectangle is removed from the calculated first circumscribed rectangle, and the remaining first circumscribed rectangle is replaced with the third circumscribed rectangle. A shape analysis step that is generated by considering it as a circumscribed rectangle;
An overlapping region removing step of generating the third circumscribed rectangle of the third circumscribed rectangle in the inclusion relation by removing the included third circumscribed rectangle and considering the remaining third circumscribed rectangle as the fourth circumscribed rectangle;
Of the fourth circumscribed rectangle, the ease of binarization of the circumscribed rectangular area, the small dispersion of the stroke width, the maximum stroke width, the first value pixel area and the second value after binarization The complexity of the boundary with the pixel area and the maximum stroke length are each digitized, and those that are determined to be non-character circumscribed rectangle areas are removed from the fourth circumscribed rectangle, and the remaining fourth circumscribed rectangle is replaced with the fourth circumscribed rectangle. A character-likeness analysis step that is generated by considering it as 5 circumscribed rectangles;
A character graphic string region output step for outputting the fifth circumscribed rectangle as a character graphic string region;
A character graphic string extraction method characterized by comprising:   A character graphic string extraction program characterized in that the character graphic string extraction method according to claim 6 is described as a computer program executable by a computer.   7. A recording medium, wherein the character graphic string extraction method according to claim 6 is described in a computer-executable program and the program is recorded.