JP2016086273A - Binarization method of image, program, and binarizing device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To realize binarization which substantially prevents blur or crush.SOLUTION: A binarizing device detects a pixel, as a ho horizontal peak pixel, having an extremal value when an input image of a multiple value is scanned along in a horizontal direction. A first to a forth differences (DIF, DIF, DIF, and DIF) of a pixel value (v[x,y]) of the horizontal peak pixel and the other reference pixel value (v[x-1,y], v[x+1,y], v[x-2,y], and v[x+2,y]) are calculated with reference to a pixel value of continuous five pixels in the horizontal direction including the horizontal peak pixel as a center. A sum of a left side edge intensity based on the difference (DIFand DIF) in the left side of horizontal peak pixel and a right edge intensity based on the difference (DIFand DIF) of the right side of horizontal peak pixel is obtained as a horizontal peak intensity. In the case where an absolute value of the horizontal peak intensity is equal to a predetermined value or more, the horizontal peak pixel is binarized into 0 or 1 in accordance with a code of the horizontal peak intensity, and the same process is performed in a vertical direction.SELECTED DRAWING: Figure 7

Description

  The present invention relates to a binarization method for binarizing a multilevel image read by a scanner or the like, a program for realizing the binarization method, and a binarization apparatus.

  In the binarization processing, each pixel of the input image is binarized by comparing a multi-value input image (grayscale image) captured via a scanner or the like with a certain threshold value, thereby obtaining a binarized image. For example, a pixel having a pixel value less than the threshold is determined to be black (binarization with a black pixel value), and a pixel having a pixel value equal to or greater than the threshold is determined to be white (the white pixel value is And binarize). A predetermined fixed value may be used as a threshold value, or a threshold value may be set using a discriminant analysis method or the like based on an input image.

  Various threshold setting methods for optimizing binarization have been proposed, but due to insufficient resolution of the scanner etc., it is judged that the thin black line part is more than the threshold and is white, The white part adjacent to the black region may be determined to be black because it is less than the threshold value. The former is called “blurred” and the latter is called “collapsed”.

  Focusing on the fact that the pixel that can be the target of crushing or blurring is usually a ridge having a lighter density or a valley having a higher density than the neighboring pixels, and in Patent Document 1, it is a ridge or a valley compared to the neighboring pixels. Pixels are detected and binarization is performed without using a threshold for detected pixels, while binarization is performed for non-detected pixels using a threshold. In Patent Document 1, when detecting whether a pixel of interest corresponds to a pixel of a ridge or a valley, only the pixel adjacent to the pixel of interest (FIG. 7 of Patent Document 1) or a position separated by one pixel from the pixel of interest Only the pixels to be processed (FIG. 8 or FIG. 9 of Patent Document 1) are referred to.

JP-A-5-284356

  By the method of Patent Document 1, blurring and crushing can be solved to some extent in the binarization process. That is, in the binarization process, it is possible to determine a faint portion as black and a collapsed portion as white to some extent. However, there is room for improvement in the method (details will be described later).

  Therefore, an object of the present invention is to provide an image binarization method, a program, and a binarization apparatus that contribute to the elimination of blurring and crushing.

  A binarization method according to the present invention is a binarization method for binarizing a multi-valued input image, and a pixel having an extreme value when the input image is scanned along a predetermined direction is defined as a peak pixel. Extracting the peak pixel based on a pixel value of each pixel in an evaluation pixel group including a peak pixel extracting step to extract and five pixels arranged continuously along the predetermined direction and including the peak pixel as a central pixel; And a binarization step for binarization.

  Specifically, for example, in the binarization method, the evaluation pixel group includes a first pixel and a second pixel located on both sides of the peak pixel in the predetermined direction, and the peak pixel in the predetermined direction. A third pixel located outside the first pixel and a fourth pixel located outside the second pixel, wherein the first pixel is located between the peak pixel and the third pixel, and the second pixel The pixel is located between the peak pixel and the fourth pixel, and in the binarization step, based on first to fourth differences in pixel values between the peak pixel and the first to fourth pixels, It is preferable to binarize the peak pixel.

  More specifically, for example, the binarization method further includes a peak intensity derivation step of deriving a sum of the first edge intensity and the second edge intensity in the peak pixel as the peak intensity of the peak pixel, and the peak intensity In the derivation step, when the first difference and the third difference have the same sign or the first difference is zero and the absolute value of the third difference is larger than the absolute value of the first difference, Deriving a third difference as the first edge strength, otherwise, deriving the first difference as the first edge strength, and the second difference and the fourth difference have the same sign or When the second difference is zero and the absolute value of the fourth difference is greater than the absolute value of the second difference, the fourth difference is derived as the second edge strength, while when not, 2 differences Derives a serial second edge intensity, the binarization process, when the absolute value of the peak intensity is a predetermined value or more, may binarizing said peak pixel based on the sign of the peak intensity.

  For example, in the binarization step, the peak pixels whose absolute value of the peak intensity is less than the predetermined value and pixels other than the peak pixels may be binarized using a threshold value.

  Further, for example, the threshold value may be a fixed value set in advance, or may be set based on the input image.

  And it is good to form the program for making a computer implement | achieve the said binarization method.

  The binarization apparatus according to the present invention is a binarization apparatus that binarizes a multi-value input image, the input image acquisition unit acquiring the input image, and the input image along a predetermined direction. A peak pixel extraction unit that extracts a pixel that takes an extreme value when scanned as a peak pixel, and an evaluation pixel group that includes five pixels arranged continuously along the predetermined direction and includes the peak pixel as a central pixel. A binarization unit that binarizes the peak pixel based on a pixel value of each pixel.

  According to the present invention, since the peak pixel is binarized based on the pixel value of each pixel in the evaluation pixel group including five consecutive pixels, it is possible to consider in detail the change in shading near the peak pixel. Therefore, it is possible to realize a good binarization in which fading or crushing is difficult to occur (in other words, fading or crushing is eliminated).

It is a schematic whole block diagram of the character recognition apparatus which concerns on 1st Embodiment which concerns on 1st Embodiment of this invention. FIG. 4 is a diagram for defining an axis, a direction, and a pixel position related to a two-dimensional image according to the first embodiment of the present invention. It is a figure which concerns on 1st Embodiment of this invention and shows the relationship (a) of the pixel value and density in an input image, and the relationship (b) of the pixel value and color in a binarized image. It is a figure which concerns on 1st Embodiment of this invention and shows the example of the light / dark distribution of an input image. It is a flowchart of the binarization process which concerns on 1st Embodiment of this invention. It is a flowchart of the binarization process which concerns on 1st Embodiment of this invention. It is a figure for demonstrating the derivation | leading-out method of horizontal edge intensity | strength concerning 1st Embodiment of this invention. It is a figure for demonstrating the classification | category conditions to a collapse candidate pixel or a blurred candidate pixel regarding 1st Embodiment of this invention in relation to a horizontal peak intensity | strength. It is a figure for demonstrating the classification conditions to a collapse candidate pixel or a blur candidate pixel regarding 1st Embodiment of this invention in relation to a vertical peak intensity | strength. It is a figure which concerns on 1st Embodiment of this invention and shows the specific example of binarization. It is a figure which shows the specific example to which the contradiction resolution process of 1st Embodiment of this invention is applied. It is a partial block diagram of the character recognition apparatus which concerns on 2nd Embodiment of this invention. It is explanatory drawing of the character recognition procedure which concerns on 2nd Embodiment of this invention. It is a functional block diagram of the binarized image generation part of FIG.

  Hereinafter, an example of an embodiment of the present invention will be specifically described with reference to the drawings. In each of the drawings to be referred to, the same part is denoted by the same reference numeral, and redundant description regarding the same part is omitted in principle. In this specification, for simplification of description, the names of information, signals, physical quantities, members, etc. corresponding to the symbols or signs are indicated by writing symbols or signs referring to information, signals, physical quantities, members, etc. Omitted or abbreviated.

<< First Embodiment >>
A first embodiment of the present invention will be described. FIG. 1 is a schematic overall block diagram of a character recognition device 1 according to the first embodiment. The character recognition device 1 includes an input image acquisition unit 10, a binarized image generation unit 20, a character recognition processing unit 30, and an output unit 40.

  The input image acquisition unit 10 acquires an input image via a camera, a scanner, or a communication line (Internet network or the like). The input image is a two-dimensional image including arbitrary characters. In detail, the image including characters means an image including image information about characters. The two-dimensional image including characters may be, for example, an image of a form on which handwritten or printed characters are described. The input image acquisition unit 10 also functions as an input image storage unit that stores the acquired input image.

  The binarized image generation unit 20 generates a binarized image by binarizing the input image. The character recognition processing unit 30 recognizes characters included in the input image by applying a known character recognition process to the binarized image. The output unit 40 outputs a character recognition result by the character recognition processing unit 30. The character recognition result by the character recognition processing unit 30 may be output as data to a memory (not shown) connected to or included in the character recognition device 1, or displayed as a video connected to or included in the character recognition device 1. It may be output to a device (not shown) or may be output to an external device (not shown) of the character recognition device 1 via a communication line. The input image acquisition unit 10, the binarized image generation unit 20, the character recognition processing unit 30, and the output unit 40 are a calculation unit (not shown) that performs binarized image generation processing, character recognition processing, and control, and binary processing. A memory (not shown) in which a program for performing a digitized image generation process, a character recognition process, and control is stored, and a storage unit (not shown) in which image data and calculation data are stored.

  As shown in FIG. 2, an arbitrary two-dimensional image such as an input image or a binarized image is formed by arranging a plurality of pixels along each of the X axis and Y axis directions orthogonal to each other. It is assumed that the X axis is parallel to the horizontal direction of the two-dimensional image and the Y axis is parallel to the vertical direction of the two-dimensional image. When considering directions for a two-dimensional image, it is assumed that the horizontal direction is parallel to the left-right direction and the vertical direction is parallel to the up-down direction. A certain pixel in the two-dimensional image is represented by a symbol P [x, y]. In the two-dimensional image, pixels arranged at positions separated by i pixels on the right, left, lower, and upper sides from the pixel P [x, y] are represented by symbols P [x + i, y] and P [xi, respectively. , Y], P [x, y + i], and P [x, y−i] (i is an integer).

  The input image is a multi-valued image in which pixel values are quantized in three or more stages. That is, the pixel value of each pixel forming the input image takes one of three or more digital values. Here, it is assumed that the input image is a grayscale image expressed by 8 bits. Therefore, the pixel value of each pixel of the input image takes an integer value of 0 or more and 255 or less. As shown in FIG. 3A, in a certain pixel in the input image, when the pixel value is 0, the density of the pixel is the highest (and therefore the luminance is the lowest), and the pixel value increases from 0 to 255. It is assumed that the density of the pixel becomes lighter as the time goes on (thus, the luminance increases). In the binarized image, each pixel is classified into a black pixel (pixel having black color) or a white pixel (pixel having white color). As shown in FIG. 3B, in a certain pixel in the binarized image, if the pixel value is 0, the pixel is a white pixel, and if the pixel value is 1, the pixel is a black pixel. In the input image and the binarized image, the relationship between the pixel value and the density is not limited to that assumed here.

  When binarizing an input image, blurring or collapse may be a problem. Originally, it should be determined as a black pixel, but a pixel that can be determined as a white pixel due to a relatively low density in the periphery and insufficient resolution of the scanner corresponds to a faint portion. On the other hand, a pixel that should be originally determined as a white pixel, but a pixel that can be determined as a black pixel due to a relatively high density in the periphery and insufficient resolution of the scanner, corresponds to a collapsed portion.

  FIG. 4 shows an example of the light and shade distribution of the input image including the character “job”. In FIG. 4, the density distribution (pixel value distribution) when the pixel values of the input image are scanned along the A line parallel to the X axis, and the pixels of the input image along the B line parallel to the Y axis. A shade distribution (pixel value distribution) when a value is scanned is shown. In FIG. 4, the white circle portion corresponds to a collapsed portion that should be determined as white, and the black circle portion corresponds to a blurred portion that should be determined as black.

  In this embodiment, in the process of generating a binarized image, it is estimated whether each pixel of the input image can be a faint or collapsed pixel, and the binarized image that should be originally used is obtained by using the estimation result. Is generated. For example, a pixel that is estimated to be a faint pixel (corresponding to a faint candidate pixel that will be described later) is binarized to a pixel of “1”, that is, a black pixel, and a pixel that is estimated to be a crushed pixel (a shadow candidate pixel that will be described later) 2) is binarized into “0” pixels, that is, white pixels.

[flowchart]
5 and 6 are flowcharts of the binarization process for generating the binarized image from the input image. A binarization process is executed by the binarized image generation unit 20. In the binarization process, first, the processes in steps S11 to S17 in FIG. 5 are sequentially performed.

  In step S <b> 11, the binarized image generation unit 20 detects a pixel whose pixel value is an extreme value when the input image is scanned along the horizontal direction (in other words, the pixel value is extremely high in the horizontal direction of the input image). A pixel having a value) is extracted from the input image as a horizontal peak pixel. That is, for example, regarding the pixel value v of a certain pixel of interest P [x, y] in the input image, “v [x−1, y] <v [x, y]” and “v [x, y]> v When [x + 1, y] ”is satisfied, or when“ v [x−1, y]> v [x, y] ”and“ v [x, y] <v [x + 1, y] ”are satisfied. The target pixel P [x, y] is extracted as a horizontal peak pixel. Here, v [x, y] represents the pixel value of the pixel P [x, y] in the input image. Therefore, for example, v [x−1, y] and v [x + 1, y] are respectively the pixel values of the pixel P [x−1, y] adjacent to the left of the pixel P [x, y] in the input image, The pixel value of the pixel P [x + 1, y] adjacent to the right of the pixel P [x, y] is represented.

  Even if the pixel value of the pixel of interest P [x, y] and the adjacent pixel P [x-1, y] and / or P [x + 1, y] of the pixel of interest P [x, y] are the same, The pixel P [x, y] can be a horizontal peak pixel. For example, even if “v [x−1, y] = v [x, y]”, “v [x−2, y] <v [x, y]” and “v [x, y]> v [x + 1, y] ”holds or“ v [x−2, y]> v [x, y] ”and“ v [x, y] <v [x + 1, y] ”hold The pixel of interest P [x, y] may be extracted as a horizontal peak pixel.

  In step S12, the binarized image generation unit 20 derives the horizontal peak intensity for each horizontal peak pixel. However, when this horizontal peak intensity is derived, the pixel values of the left and right pixels of the horizontal peak pixel are referred to, and the sum of the left and right edge intensity is set as the horizontal peak intensity.

  With reference to FIGS. 7A and 7B, assuming that the horizontal peak pixel is the pixel P [x, y], the method for deriving the horizontal peak intensity in step S12 will be specifically described.

First, the binarized image generation unit 20 uses the horizontal peak pixel P [x, y] as a reference, the pixel value difference DIF _L1 between the pixels P [x, y] and P [x−1, y], the pixel A pixel value difference DIF _L2 between P [x, y] and P [x−2, y], a pixel value difference DIF _R1 between pixels P [x, y] and P [x + 1, y], and a pixel A pixel value difference DIF _R2 between P [x, y] and P [x + 2, y] is obtained. Differences DIF _L1 and DIF _L2 indicate the edge strength on the left side of horizontal peak pixel P [x, y], while differences DIF _R1 and DIF _R2 indicate the edge strength on the right side of horizontal peak pixel P [x, y]. They are represented by the following formulas (A1) to (A4).
DIF _L2 = v [x, y] −v [x−2, y] (A1)
DIF _L1 = v [x, y] −v [x−1, y] (A2)
DIF _R1 = v [x, y] −v [x + 1, y] (A3)
DIF _R2 = v [x, y] −v [x + 2, y] (A4)

Subsequently, the binarized image generation unit 20 determines that the difference DIF _L1 and the difference DIF _L2 have the same sign (same polarity) or the difference DIF _L1 is zero, and the absolute value of the difference DIF _L2 is the difference DIF. When larger than the absolute value of _L1 , the difference DIF _L2 is obtained as the left edge strength, and when not, the difference DIF _L1 is obtained as the left edge strength. Further, the binary image generation unit 20 has the difference DIF _R1 and the difference DIF _R2 of the same sign (same polarity) or the difference DIF _R1 is zero, and the absolute value of the difference DIF _R2 is the difference DIF _R1. Is greater than the absolute value of DIF _R2 , the difference DIF _R2 is determined as the right edge strength, otherwise the difference DIF _R1 is determined as the right edge strength. Then, the binarized image generation unit 20 uses the sum of the left edge intensity and the right edge intensity obtained for the horizontal peak pixel P [x, y] as the horizontal edge intensity of the horizontal peak pixel P [x, y]. To derive.

Therefore, for example, as shown in FIG. 7A, “v [x−2, y]> v [x−1, y]> v [x, y]” and “v [x, y] <v [x + 1”. , Y] <v [x + 2, y] ”, the horizontal edge intensity of the horizontal peak pixel P [x, y] is“ DIF _L2 + DIF _R2 ”. Or, for example, as shown in FIG. 7B, “v [x−2, y]> v [x−1, y]> v [x, y]” and “v [x, y] <v [x + 1”. , Y] ”and“ v [x + 1, y]> v [x + 2, y] ”, the horizontal edge intensity of the horizontal peak pixel P [x, y] is“ DIF _L2 + DIF _R1 ”.

  In step S13 subsequent to step S12, the binarized image generation unit 20 executes candidate determination processing for each horizontal peak pixel. In the candidate determination process of step S13, the horizontal peak pixel is any one of the blurred candidate pixel, the collapse candidate pixel, and the non-candidate pixel based on the sign (polarity) of the horizontal peak intensity of the horizontal peak pixel and the absolute value of the horizontal peak intensity. Classify into:

Assuming that the horizontal peak pixel is the pixel P [x, y], the candidate determination process in step S13 will be specifically described. Please refer to FIG. In the case CASE _H1 in which the horizontal peak intensity of the horizontal peak pixel P [x, y] is negative and equal to or less than the negative predetermined value TH _{HL, the} binarized image generation unit 20 performs the horizontal peak pixel P [x, y] is classified as a blurred candidate pixel. In CASE _H2 in which the horizontal peak intensity of the horizontal peak pixel P [x, y] is positive and is equal to or greater than a predetermined positive value TH _HU , the horizontal peak pixel P [x, y] is classified as a collapse candidate pixel, and in case CASE _H3 , which does not apply to any of the cases CASE _H1 and CASE _H2 , the horizontal peak pixel P [x, y] is classified as a non-candidate pixel. The absolute value of the predetermined value TH _{HL and} the absolute value of the predetermined value TH _HU are the same, but may be different from each other.

  A pixel classified as a blurred candidate pixel is estimated to be a pixel with a high possibility of blurring, and is binarized with a pixel value of “1” in principle (see step S25 in FIG. 6). Pixels classified as collapse candidate pixels are presumed to be pixels that are highly likely to be collapsed, and are later binarized with a pixel value of “0” in principle (see step S26 in FIG. 6). For this reason, if the pixel of interest is classified as a collapse candidate pixel or a collapse candidate pixel simply because the pixel of interest has an extreme value in the horizontal direction, the pixel is easily affected by noise or the like. For this reason, in step S13, only when the absolute value of the horizontal peak intensity is greater than or equal to a predetermined value, the target pixel is classified as a blurred candidate pixel or a collapsed candidate pixel. The same applies to step S16 described later.

  In step S <b> 14, the binarized image generation unit 20 determines whether the pixel value has an extreme value when the input image is scanned along the vertical direction (in other words, the pixel value has an extreme value in the vertical direction of the input image). A pixel having a value) is extracted from the input image as a vertical peak pixel. That is, for example, regarding the pixel value v of a certain pixel of interest P [x, y] in the input image, “v [x, y−1] <v [x, y]” and “v [x, y]> v When [x, y + 1] ”is satisfied, or when“ v [x, y−1]> v [x, y] ”and“ v [x, y] <v [x, y + 1] ”are satisfied. The target pixel P [x, y] is extracted as a vertical peak pixel.

  Even if the pixel value of the pixel of interest P [x, y] and the adjacent pixel P [x, y-1] and / or P [x, y + 1] of the pixel of interest P [x, y] are the same, The pixel P [x, y] can be a vertical peak pixel. For example, even if “v [x, y−1] = v [x, y]”, “v [x, y−2] <v [x, y]” and “v [x, y]> When v [x, y + 1] ”is satisfied or when“ v [x, y−2]> v [x, y] ”and“ v [x, y] <v [x, y + 1] ”are satisfied. The pixel of interest P [x, y] may be extracted as a vertical peak pixel.

  In step S15, the binarized image generation unit 20 derives the vertical peak intensity for each vertical peak pixel. However, when deriving the vertical peak intensity, the pixel values of the upper and lower pixels of the vertical peak pixel are referred to, and the sum of the upper and lower edge intensity is set as the vertical peak intensity.

  Assuming that the vertical peak pixel is the pixel P [x, y], the method for deriving the vertical peak intensity in step S15 will be specifically described.

First, the binarized image generation unit 20 uses the vertical peak pixel P [x, y] as a reference, the pixel value difference DIF _U1 between the pixels P [x, y] and P [x, y−1], the pixel The pixel value difference DIF _U2 between P [x, y] and P [x, y-2], the pixel value difference DIF _D1 between the pixels P [x, y] and P [x, y + 1], and the pixel A pixel value difference DIF _D2 between P [x, y] and P [x, y + 2] is obtained. Differences DIF _U1 and DIF _U2 indicate the edge strength above the vertical peak pixel P [x, y], while differences DIF _D1 and DIF _D2 indicate the edge strength below the vertical peak pixel P [x, y]. These are represented by the following formulas (B1) to (B4).
DIF _U2 = v [x, y] -v [x, y-2] (B1)
DIF _U1 = v [x, y] −v [x, y−1] (B2)
DIF _D1 = v [x, y] −v [x, y + 1] (B3)
DIF _D2 = v [x, y] -v [x, y + 2] (B4)

Subsequently, the binarized image generation unit 20 determines that the difference DIF _U1 and the difference DIF _U2 have the same sign (same polarity) or the difference DIF _U1 is zero and the absolute value of the difference DIF _U2 is the difference DIF. _When the absolute value of _U1 is larger, the difference DIF _U2 is obtained as the upper edge strength, and when not, the difference DIF _U1 is obtained as the upper edge strength. Further, the binary image generation unit 20 has the difference DIF _D1 and the difference DIF _D2 having the same sign (same polarity) or the difference DIF _D1 is zero, and the absolute value of the difference DIF _D2 is the difference DIF _D1. Is greater than the absolute value of the difference DIF _D2 as the lower edge strength, otherwise the difference DIF _D1 is determined as the lower edge strength. Then, the binarized image generation unit 20 uses the sum of the upper edge intensity and the lower edge intensity obtained for the vertical peak pixel P [x, y] as the vertical edge intensity of the vertical peak pixel P [x, y]. Derived as

  In step S16 subsequent to step S15, the binarized image generation unit 20 performs candidate determination processing for each vertical peak pixel. In the candidate determination processing in step S16, the vertical peak pixel is any one of the blurred candidate pixel, the collapse candidate pixel, and the non-candidate pixel based on the sign (polarity) of the vertical peak intensity of the vertical peak pixel and the absolute value of the vertical peak intensity. Classify into:

Assuming that the vertical peak pixel is the pixel P [x, y], the candidate determination process in step S16 will be specifically described. Please refer to FIG. In the case CASE _V1 in which the vertical peak intensity of the vertical peak pixel P [x, y] is negative and is equal to or less than the negative predetermined value TH _{VL, the} binarized image generation unit 20 performs the vertical peak pixel P [x, y] are classified as fading candidate pixels. In CASE _V2 in which the vertical peak intensity of the vertical peak pixel P [x, y] is positive and equal to or greater than a predetermined positive value TH _VU , the vertical peak pixel P [x, y] is classified as a collapse candidate pixel, and in case CASE _V3 , which does not apply to either case CASE _{V1 or} CASE _V2 , the vertical peak pixel P [x, y] is classified as a non-candidate pixel. The absolute value of the predetermined value TH _{VL and} the absolute value of the predetermined value TH _VU are the same, but may be different from each other. The predetermined value TH _HL (see FIG. 8) and the predetermined value TH _VL (see FIG. 9) are the same, but may be different from each other. The predetermined value TH _HU (see FIG. 8) and the predetermined value TH _VU (see FIG. 9) are the same, but may be different from each other.

  The left edge intensity used for the calculation of the horizontal peak intensity is also called a first edge intensity, and the right edge intensity is also called a second edge intensity. Further, the upper edge strength used for calculating the vertical peak strength is also referred to as a first edge strength, and the lower edge strength is also referred to as a second edge strength.

  In the flowchart of FIG. 5, the processes of steps S14 to S16 are executed after the processes of steps S11 to S13. However, the processes of steps S14 to S16 are executed before the processes of steps S11 to S13. Alternatively, the processes in steps S11 to S13 and the processes in steps S14 to S16 may be executed simultaneously. Further, when calculating the difference using the expressions (A1) to (A4) and the expressions (B1) to (B4), the neighboring pixels are calculated from the pixel value v [x, y] of the reference peak pixel P [x, y]. However, the pixel value v [x, y] of the peak pixel P [x, y] may be subtracted from the pixel value of the neighboring pixel. . However, in that case, in steps S13 and S16, it is necessary to reverse the classification method into the blurred candidate pixel and the collapsed candidate pixel.

  In step S <b> 17, the binarized image generation unit 20 classifies pixels that are neither horizontal peak pixels nor vertical peak pixels among the pixels of the input image as non-candidate pixels.

  After step S17, the process proceeds to step S21 (see FIG. 6). In the processing after step S21, in principle, the pixels classified as the blurred candidate pixels are binarized into “1” pixels, that is, black pixels (steps S25 and S28), and the pixels classified as the collapsed candidate pixels. Is binarized into “0” pixels, that is, white pixels (steps S26 and S30). However, since the classification results in steps S13 and S16 may conflict with each other, a measure for the contradiction is also prepared (step S31). Hereinafter, the processes of steps S21 to S34 executed by the binarized image generation unit 20 will be described in detail.

  In step S <b> 21, the binarized image generation unit 20 sets all pixels classified as a blurred candidate pixel or a collapsed candidate pixel as first to Nth candidate pixels. N represents the total number of pixels classified as a blurred candidate pixel or a collapsed candidate pixel.

  A single pixel P [x, y] is extracted as a horizontal peak pixel and classified as a blurred candidate pixel or a collapsed candidate pixel in step S13, and is extracted as a vertical peak pixel and blurred candidate pixel in step S16. Or it may be classified as a collapse candidate pixel. The single pixel P [x, y] subjected to such classification does not occupy two candidate pixels (for example, the first and second candidate pixels) among the first to Nth candidate pixels, One candidate pixel (for example, the first candidate pixel) among the first to Nth candidate pixels is set. Therefore, when such classification is performed, the total number of horizontal peak pixels classified into the blurred candidate pixel or the collapsed candidate pixel in Step S13 and the vertical classified as the blurred candidate pixel or the collapsed candidate pixel in Step S16. The sum of the total number of peak pixels is larger than N.

  In step S22 following step S21, 1 is assigned to the variable j, and then the process proceeds to step S23. In step S <b> 23, the binarized image generation unit 20 determines whether or not the jth candidate pixel is a horizontal peak pixel and a vertical peak pixel. If the jth candidate pixel is a horizontal peak pixel and a vertical peak pixel, the process proceeds to step S27. Otherwise, the jth candidate pixel is either the horizontal peak pixel or the vertical peak pixel. The process proceeds to step S24.

  In step S24, the binarized image generation unit 20 determines whether or not the jth candidate pixel is classified as a blurred candidate pixel. If the jth candidate pixel is classified as a blurred candidate pixel, step S25 is performed. On the other hand, if not, that is, if the j-th candidate pixel is classified as a collapsed candidate pixel, the process proceeds to step S26. In step S <b> 25, the binarized image generation unit 20 sets a pixel value of “1” to the jth candidate pixel. In step S <b> 26, the binarized image generation unit 20 sets a pixel value of “0” to the jth candidate pixel. After step S25 or S26, the process proceeds to step S32.

  By setting a pixel value of “1” or “0” to the j-th candidate pixel, the j-th candidate pixel is binarized (the same applies to steps S28, S30, and S31 described later). Setting the pixel values of “1” and “0” for the jth candidate pixel means that “1” and “0” are respectively set to the pixel values of the pixels in the binarized image corresponding to the jth candidate pixel. It means to assign. In other words, setting the pixel values of “1” and “0” for the jth candidate pixel means that the pixels in the binarized image corresponding to the jth candidate pixel are set to black pixels and white pixels, respectively. It means to do.

  In step S27, the binarized image generation unit 20 determines whether or not the jth candidate pixel has been classified as a blurred candidate pixel in steps S13 and S16. When the jth candidate pixel is classified as a blurred candidate pixel in steps S13 and S16, the binarized image generation unit 20 sets a pixel value of “1” to the jth candidate pixel in step S28. The process proceeds from step S32 to step S32. If not, the process proceeds from step S27 to step S29.

  In step S29, the binarized image generation unit 20 determines whether or not the jth candidate pixel has been classified as a collapsed candidate pixel in steps S13 and S16. When the jth candidate pixel is classified as a collapsed candidate pixel in steps S13 and S16, the binarized image generation unit 20 sets a pixel value of “0” to the jth candidate pixel in step S30. From step S32, the process proceeds to step S31. Otherwise, the process proceeds to step S31.

  When the classification results of steps S13 and S16 for the common pixel are inconsistent with each other, the process reaches step S31 for the jth candidate pixel as the common pixel. That is, the j-th candidate pixel, which is a single common pixel, is extracted as a horizontal peak pixel and classified as a blurred candidate pixel in step S13, and is extracted as a vertical peak pixel and becomes a collapsed candidate pixel in step S16. If classified, or the jth candidate pixel, which is a single common pixel, is extracted as a horizontal peak pixel and classified as a collapsed candidate pixel in step S13, and extracted as a vertical peak pixel in step S16. If the pixel is classified as a blurred candidate pixel, the process proceeds to step S31.

  In step S31, the binarized image generation unit 20 sets a pixel value of “1” or “0” to the jth candidate pixel by the contradiction resolution process, and then proceeds to step S32.

For the sake of specific description, the contradiction resolution process will be described on the assumption that the jth candidate pixel is the pixel P [x, y]. In the contradiction resolution process, the absolute value of the left edge strength based on the above-mentioned difference DIF _L1 and DIF _L2 (hereinafter represented by the symbol | EG _L |) and the absolute value of the right edge strength based on the above-described difference DIF _R1 and DIF _R2 A value (hereinafter represented by the symbol | EG _R |), an absolute value of the upper edge intensity (hereinafter represented by the symbol | EG _U |) based on the above-described differences DIF _U1 and DIF _U2 , and the above-described difference DIF _D1 And the absolute value of the lower edge strength based on DIF _D2 (hereinafter represented by the symbol | EG _D |). The absolute values | EG _L |, | EG _R |, | EG _U |, and | EG _D | are obtained for the pixel P [x, y] as the j-th candidate pixel, and the difference DIF _L1 etc. Is calculated according to the above formulas (A1) to (A4) and formulas (B1) to (B4). The calculation method of the left side, right side, upper side, and lower side edge strength is also as described above.

In the contradiction resolution process, the binarized image generation unit 20 binarizes the jth candidate pixel by giving priority to the classification result in step S16 over the classification result in step S13 if the following formula (C1) is satisfied. Otherwise, the jth candidate pixel is binarized with priority given to the classification result of step S13 over the classification result of step S16. In the formula (C1), the left side represents the smaller absolute value of | EG _U | and | EG _D |, and the right side represents the smaller absolute value of | EG _L | and | EG _R |. Represent. By not adopting the result in the direction where the absolute value of the edge intensity is the smallest, that is, the direction where the pixel value is closest, the determination of crushing or fading is performed.
Min (| EG _U |, | EG _D |)> Min (| EG _L |, | EG _R |)
... (C1)

  Therefore, for example, when the pixel P [x, y] as the jth candidate pixel is classified as a blurred candidate pixel in step S13, but is classified as a collapsed candidate pixel in step S16, Expression (C1) is established. Then, the pixel P [x, y] is finally classified as a collapsed candidate pixel, while if the expression (C1) is not established, the pixel P [x, y] is finally classified as a blurred candidate pixel. Alternatively, for example, when the pixel P [x, y] as the jth candidate pixel is classified as a collapse candidate pixel in step S13, but is classified as a blur candidate pixel in step S16, Expression (C1) is established. Then, the pixel P [x, y] is finally classified as a blurred candidate pixel, while if the expression (C1) is not established, the pixel P [x, y] is finally classified as a collapsed candidate pixel.

  In the contradiction resolution processing, when the pixel P [x, y] is finally classified as a candidate for squashing, a pixel value of “0” is set to the jth candidate pixel, and the pixel P [x, y] is final. When the pixel is classified as a faint candidate pixel, a pixel value of “1” is set to the jth candidate pixel.

  In step S32, it is determined whether or not the variable j matches the value N. If “j = N” is satisfied, the process proceeds to step S34. If “j = N” is not satisfied, the process proceeds to step S33. After adding 1 to the variable j, the process returns to step S23 to repeat the processes after step S23.

  In step S34, the binarized image generation unit 20 binarizes each pixel classified as a non-candidate pixel in steps S13, S16, and S17 using a predetermined threshold th. That is, for each non-candidate pixel, if the pixel value of the non-candidate pixel is equal to or greater than the threshold th, a pixel value of “0” is set to the non-candidate pixel, while if the pixel value of the non-candidate pixel is less than the threshold th, the non-candidate pixel A pixel value of “1” is set in. Any known binarization method can be used as the binarization method in step S34. Therefore, for example, the threshold th used in step S34 may be a fixed value set in advance, or may be set using a discriminant analysis method based on the pixel value of each non-candidate pixel, It may be set using the method described in Japanese Patent No. 3831797 based on the pixel value of each non-candidate pixel. The binarized image generation unit 20 can include a threshold setting unit (not shown) that sets the threshold th based on the pixel value of each non-candidate pixel.

  Since all the pixels of the input image are binarized by the end of the process of step S34, the binarization process is finished.

  In the recognition of characters through binarization, when trying to recognize small characters, it may not be correctly recognized due to crushing or blurring, but in this embodiment, it is possible to realize binarization that does not easily cause blurring or blurring. Even small characters can be recognized accurately. In addition, since squashing can be eliminated and binarization can be performed, a beneficial effect is also exhibited in separating characters that are close enough to contact each other.

[Specific binarization example]
Next, a specific example of an image is shown, and an example of binarization is shown. Referring to FIG. 10A, as a simple model, a paper surface 310 on which three thin first to third black lines 311 to 313 are drawn in parallel and closely to each other is prepared, and the drawing contents on the paper surface 310 are prepared. Let us consider a situation in which an input image 320 is obtained by capturing the image with a scanner. The black lines 321 to 323 correspond to the first to third black lines on the input image 320, respectively. Each of the first to third black lines is a black line segment extending in the vertical direction, and the first to third black lines are arranged in the left-right direction. Further, as shown in FIG. 10B, at the time of capturing by the scanner, each of the black lines 311 to 313 has a width corresponding to one pixel, and the center interval between the black lines 311 and 312 and the black lines 312 and 313. Assume that both of the center intervals have a width of 1.5 pixels. In FIG. 10B, a square with a symbol P _SCAN represents a pixel at the time of capturing by the scanner (however, only a part of the pixels has a symbol P _SCAN ). Further, it is assumed that the center of a specific pixel 315 is located on the center line of the black line 311 at the time of capturing of the scanner. As a result, 1, 2, 3, 4, Pixels located to the right by 5, 6 and 7 pixels respectively capture white, gray, gray, white, black, white and white shading information and are located on the left side with respect to a specific pixel 315 All of the pixels that have been taken in will capture the shade information of white. Pixels that will capture gray shade information are pixels that are located on the black line 312 by half of the pixels.

  Then, an enlarged view of the black lines 321 to 323 on the input image 320 is as shown in FIG. 10C (in FIG. 10B and FIG. 10C, the color density is changed by the hatched diagonal density. expressing). The pixel P [x−2, y] corresponds to the specific pixel 315 described above. Therefore, the center of the pixel P [x−2, y] is located at the center of the black line 321, and in the input image 320, the density information of the pixels P [x−4, y] to P [x + 5, y] is , White, white, black, white, gray, gray, white, black, white, white, respectively. If such an input image 320 is binarized using a general threshold, if the pixel values of the pixels P [x, y] and P [x + 1, y] are equal to or greater than the threshold, the pixel P [x, y] and P [x + 1, y] are determined to be white pixels. That is, at the positions of the pixels P [x, y] and P [x + 1, y], the black line 322 is faded and disappears by binarization. The same applies to the pixels P [x, y + 1] and P [x + 1, y + 1].

  On the other hand, according to the method of the present embodiment, since the pixel P [x, y] and / or P [x + 1, y] has a minimum value in the input image 320, depending on the edge strength of that portion, FIG. The pixels P [x, y] and / or P [x + 1, y] are blurred and classified as candidate pixels through the processes in steps S11 to S13. As a result, the pixels P [x, y] and / or P [ A pixel value of “1” is set in x + 1, y]. That is, the blur is eliminated on the binarized image.

  As methods aiming at elimination (suppression) of blurring, there are the following reference methods α1 and α2. The reference method α1 corresponds to the method using FIG. 7 in Patent Document 1, and the reference method α2 corresponds to the method using FIG. 8 or FIG. 9 in Patent Document 1. In the reference methods α1 and α2, pixels that are ridges or valleys compared to neighboring pixels are detected, and binarization is performed without using threshold values for the detected pixels, while threshold values are used for non-detected pixels. Binarization is performed using.

  However, when focusing on the left and right direction, in the reference method α1, only the left and right adjacent pixels of the target pixel are referred to as neighboring pixels. For this reason, when the pixel P [x, y] is the pixel of interest, the pixel of interest [x, y] and the neighboring pixel P [x, y + 1] are both gray, and the difference between the pixel values is a threshold value (Patent Document) The pixel of interest P [x, y] is not detected as a ridge or valley. Therefore, it is exposed to a binarization process using a conventional threshold, and as a result, depending on the threshold, the portion of the pixel P [x, y] is faded (that is, the pixel P [x, y] is converted into a white pixel). Valued). Further, in the reference direction α1, if the ridge or valley does not become a ridge or valley in two or more of the two types of oblique directions, the pixel is not detected as a ridge or valley pixel (FIG. 1). 7 and paragraphs [0016] to [0019]), also from this point, the pixel of interest P [x, y] is highly likely to be subjected to binarization processing using a conventional threshold value. The same applies to the pixel P [x + 1, y].

  When attention is paid to the left-right direction, in the reference method α2, only pixels located one pixel to the left from the target pixel and pixels located one pixel to the right from the target pixel are referred to as neighboring pixels. Therefore, when the pixel P [x, y] is the target pixel, the neighboring pixels P [x−2, y] and [x + 2, y] for the target pixel P [x, y] are black pixels and white pixels. Therefore, the pixel value difference between the target pixel P [x, y] and the neighboring pixel P [x−2, y] and the pixel between the target pixel P [x, y] and the neighboring pixel P [x + 2, y] The difference in value does not have the same sign, and the pixel of interest P [x, y] is not detected as a ridge or valley. Therefore, it is exposed to a binarization process using a conventional threshold, and as a result, depending on the threshold, the portion of the pixel P [x, y] is faded (that is, the pixel P [x, y] is converted into a white pixel). Valued). In addition, in the reference direction α2, the pixel is not detected as a pixel of a ridge or a valley unless it becomes a ridge or a valley in at least two or more directions among the vertical direction, the horizontal direction, and a plurality of oblique directions (see Patent Document 1). From FIG. 8, FIG. 9 and paragraphs [0021] to [0026]), the pixel of interest P [x, y] is highly likely to be subjected to a binarization process using a conventional threshold. The same applies to the pixel P [x + 1, y].

  The method of the present embodiment and the reference methods α1 and α2 have been described by focusing on fading, but the same can be said for the collapse.

  Next, a specific example to which the contradiction resolution process is applied will be described with reference to FIG. As a simple model, a paper surface 410 on which two thick fourth and fifth black lines 411 and 412 that are parallel and in close contact with each other are prepared, and the input image 420 is obtained by capturing the drawing content on the paper surface 410 with a scanner. Consider the situation where Black lines 421 and 422 correspond to fourth and fifth black lines on the input image 420, respectively. Each of the fourth and fifth black lines is a black line segment extending in the left-right direction, and the fourth and fifth black lines are arranged in the vertical direction. On the paper 410, the gap between the black lines 411 and 412 is about one pixel captured by the scanner. As shown in FIG. 11, the grayscale information of each pixel located in the gap between the black lines 421 and 422 in the input image 420 is gray. It has become. Also, as shown in FIG. 11, in the input image 420, it is assumed that a pixel P [x, y] exists in the vicinity of the center between the black lines 421 and 422.

  Under the assumption as described above, each pixel located in the gap between the black lines 421 and 422 is easily classified as a collapse candidate pixel in step S16 of FIG. For example, it is assumed that the pixels P [x−3, y] to P [x + 3, y] are all classified as collapse candidate pixels in step S16. However, here, the pixels P [x−3, y] to P [] are caused by the blurring or inclination of the black line 411 or 412 on the paper surface 410, the positional relationship between the scanner and the paper surface 410, noise, and the like. It is assumed that only pixel P [x, y] among x + 3, y] is also classified as a blurred candidate pixel in step S13. In this case, the contradiction resolution process is applied to the pixel P [x, y].

  At this time, since the edge strength in the horizontal direction of the pixel P [x, y] tends to be relatively smaller than the edge strength in the vertical direction, the above formula (C1) is easily established, and finally the pixel P [x , Y] are classified as collapse candidate pixels. As a result, the pixels P [x−3, y] to P [x + 3, y] are all binarized with a pixel value of “0”. When the horizontal edge strength is relatively smaller than the vertical edge strength, the change in the gray level in the horizontal direction is relatively small (because pixels having the same level of gray level information are arranged in the horizontal direction). Therefore, it is better to preferentially adopt the classification result based on the edge strength in the vertical direction (classification result in step S16) ignoring the classification result based on the edge strength in the horizontal direction (classification result in step S13). It is thought that.

  Hereinafter, several deformation methods based on the above-described binarization method will be described.

[First Modification Method]
When a certain pixel of interest P [x, y] is classified as a blurred candidate pixel in step S13 or S16 of FIG. 5, one or more neighboring pixels of the pixel of interest P [x, y] are also classified as a blurred candidate pixel. When a certain target pixel P [x, y] is classified as a collapse candidate pixel, one or more neighboring pixels of the target pixel P [x, y] are also classified as a collapse candidate pixel. Anyway. In this case, the neighboring pixels classified as the blurred candidate pixel or the collapsed candidate pixel together with the target pixel P [x, y] are deleted from the group of non-candidate pixels (that is, the process of step S34 in FIG. 6 is not applied). It is binarized to the same pixel value as the target pixel P [x, y]. Here, the one or more neighboring pixels of the target pixel P [x, y] are, for example, all or a part of the four neighboring pixels of the target pixel P [x, y] or the target pixel P [x, y]. All or part of 8 neighboring pixels.

[Second Modification Method]
In step S25 or S28 of FIG. 6, the pixel value of the i-th candidate pixel is not set unconditionally to the i-th candidate pixel that is a blurred candidate pixel, but the pixel value of the i-th candidate pixel is a predetermined threshold th _A. If this is the case, a pixel value of “0” is set for the i-th candidate pixel, while a pixel value of “1” is set for the i-th candidate pixel if the pixel value of the i-th candidate pixel is less than the threshold th _A. Also good.

Similarly, in step S26 or S30 of FIG. 6, the pixel value of the i-th candidate pixel is not set to a predetermined value for the i-th candidate pixel that is a collapse candidate pixel, but is set unconditionally. If the threshold value is equal to or greater than the threshold th _B, the pixel value of “0” is set to the i th candidate pixel, while if the pixel value of the i th candidate pixel is less than the threshold th _B , the pixel value of “1” is set to the i th candidate pixel You may do it.

Here, the thresholds th _A and th _B may be determined in advance so as to satisfy “th _A > th _B ”. As a result, even if the pixel value of the blurred candidate pixel is relatively large (even if it is a relatively light pixel; see FIG. 3A), the pixel value “1” is easily set. The occurrence of fading on the top is suppressed. Similarly, even if the collapse candidate pixel has a relatively small pixel value (even if it is a relatively dark pixel; see FIG. 3A), a pixel value of “0” is easily set. The occurrence of crushing is suppressed.

Furthermore, the thresholds th _A and th _B may satisfy “th _A >th> th _B ” in relation to the threshold th used in step S34 in FIG.

<< Second Embodiment >>
A second embodiment of the present invention will be described. The second embodiment and the third embodiment to be described later are embodiments based on the first embodiment, and matters not specifically described in the second and third embodiments are the same as those in the first embodiment unless there is a contradiction. The description also applies to the second and third embodiments.

  As shown in FIG. 12, the binarized image generation unit 20 generates a single input image IM by using different first to Mth binarization processes (in other words, M types of binarization methods). By binarizing, M binarized images may be generated. Here, M is an arbitrary integer of 2 or more. A binarized image generated using the i-th binarization process is represented by symbol IB [i]. It may be considered that the binarized image generation unit 20 includes M binarization processing units (not shown) that execute the first to Mth binarization processes.

  Here, the first binarization process is the binarization process described in the first embodiment. Therefore, the binarized image IB [1] is a binarized image obtained by using the binarization processing of FIGS. The second binarization process may be a binarization process using a known discriminant analysis method, for example. In this case, a threshold value is set using a discriminant analysis method based on the pixel value of each pixel of the input image IM, and each pixel of the input image IM is binarized using the set threshold value, whereby the binarized image IB [ 2] is obtained.

  The character recognition processing unit 30 extracts an image feature amount from the binarized image based on the pixel value of each pixel of the binarized image for each binarized image generated by the binarized image generating unit 20, A composite image feature value is generated by combining the image feature values of the binarized images IB [1] to IB [M]. Then, the character recognition processing unit 30 recognizes characters included in the input image IM based on the composite image feature amount. The composition here may be an average (simple average or weighted average).

  With reference to FIG. 13, assuming that M = 2, the operation of the character recognition processing unit 30 will be described in detail. In the example of FIG. 13, the input image IM includes kanji image information “job”, and binarized images IB [1] and IB [2] are obtained from the input image IM. The character recognition processing unit 30 generates the image feature amount IF [1] using the contour information of the binarized image IB [1] based on the pixel value of each pixel of the binarized image IB [1]. The image feature amount IF [2] is generated using the contour information of the binarized image IB [2] based on the pixel value of each pixel of the binarized image IB [2]. The image feature quantity IF [i] is composed of a plurality of dimensional quantities representing the characteristics of the binarized image IB [i]. Here, as an example, the image feature quantity IF [i] is derived from a 200-dimensional quantity representing the characteristics of the binarized image IB [i]. Shall be. Then, the image feature amount IF [i] is expressed by a 200-dimensional feature vector.

The character recognition processing unit 30 generates the composite image feature value IF _CMB by taking the average of the image feature values IF [1] and IF [2]. The average here may be a simple average or a weighted average.

  The character recognition processing unit 30 includes a template storage unit 31 that stores a plurality of templates including the template TP. Each template indicates the image feature amount of the character. A template TP shown in FIG. 13 shows an image feature amount of a Chinese character “job”. The template storage unit 31 may be provided in an external device (not shown) of the character recognition device 1, and the storage content of the storage unit 31 is supplied to the character recognition processing unit 30 via a communication line such as the Internet network. May be.

The character recognition processing unit 30 calculates the similarity between the composite image feature value IF _CMB and the image feature value of each template in the storage unit 31, and what is the character included in the input image IM based on the calculation result. Recognize. The similarity between the composite image feature amount IF _CMB and the image feature amount of a certain template is the same as the first feature vector representing the composite image feature amount IF _CMB and the second feature vector representing the image feature amount of the template. It is represented by the Euclidean distance between the first and second feature vectors when arranged in the feature space. When the Euclidean distance between the composite image feature value IF _CMB and the template TP is equal to or smaller than a predetermined reference value, the character recognition processing unit 30 determines that the character included in the input image IM corresponds to the character “job” corresponding to the template TP. ”.

  If the binarization processing of the first embodiment is used, it is possible to obtain a good character recognition result even with a faint character or a crushed character, but the best result is not always obtained in every case. For example, in principle, since it is easy to react to noise, it is possible to classify the originally white portion as a blurred candidate pixel, and it is also possible to classify a relatively lightly dark portion in the solid area as a collapsed candidate pixel. Considering this, in the second embodiment, character recognition is performed by using a composite result of a plurality of image feature amounts obtained through a plurality of binarization processes. As a result, character recognition is performed in a state where the advantages of a plurality of binarization processes are obtained (in a state in which a portion that is not good at a certain binarization process is compensated by another binarization process). Thus, a better character recognition result can be obtained than when a single binarization process is used.

  Although the example in the case of utilizing the binarization process of 1st Embodiment as a 1st binarization process was given, the 1st-Mth binarization process may be arbitrary. For example, a binarization process that gives a better result against blurring than squash is used as the first binarization process, and a binarization process that gives a better result against squashing than blurring is used as the second binarization process. If it is used as the binarization process, it becomes possible to obtain a good character recognition result for both the blurred character and the crushed character.

<< Third Embodiment >>
A third embodiment of the present invention will be described. In the third embodiment, application techniques and modification techniques applicable to the first or second embodiment will be described.

  The configuration of the binarized image generation unit 20 can be represented by the functional block diagram of FIG. The binarized image generation unit 20 of FIG. 14 includes a peak pixel extraction unit 21 that executes the processes of steps S11 and S14, a peak intensity derivation unit 22 that executes the processes of steps S12 and S15, and a binarization unit 23. . The binarization unit 23 executes the processing of steps S13, S16, and S17, the candidate binarization unit 23b that executes the processing of steps S22 to S33, and the processing of step S34. A binarization unit for candidates 23c.

  The character recognition device 1 may be realized by dedicated hardware, may be realized by executing a dedicated program (software) on a computer including the dedicated hardware, or on a general-purpose computer. It may be realized by executing a dedicated program (software). The dedicated program causes a computer to realize the binarization method described in the first or second embodiment (a method for generating a binarized image from an input image).

  In the first and second embodiments, the method of binarizing an input image including characters has been described, but the binarization method of the present invention including the binarization method described in the first and second embodiments is It is not limited to this. For example, for an input image obtained by capturing a barcode or QR code (registered trademark) with a camera or scanner, or an input image obtained by capturing a paper surface including a seal stamp with a camera or scanner, A binarization method may be applied.

  In the first embodiment, extraction of a pixel having an extreme value is performed in each of the horizontal direction and the vertical direction. However, extraction of a pixel having an extreme value is performed only in one of the horizontal direction and the vertical direction. (In this case, contradiction reduction processing is unnecessary). For example, when the occurrence of crushing or fading is regarded as a problem only in the horizontal direction, the processes of steps S14 to S16 are deleted from FIG. 5, and the processes of steps S23 and S27 to S31 are deleted from FIG. Good. In this case, after step S22 of FIG. 6, after performing the process of step S25 or S26 via step S24, the process may proceed to step S32. The same applies when the occurrence of crushing or fading is regarded as a problem only in the vertical direction. The same applies when the input image is a one-dimensional image.

<< Consideration of the Present Invention >>
The technique of the present invention embodied in the first to third embodiments will be considered.

  A binarization method W according to one aspect of the present invention is a binarization method for binarizing a multi-value input image, and a pixel that takes an extreme value when the input image is scanned along a predetermined direction. Pixel extraction step (steps S11 and S14) for extracting each pixel in the evaluation pixel group including the five pixels continuously arranged along the predetermined direction and the peak pixel as a central pixel. And a binarization step of binarizing the peak pixel based on a pixel value.

  In order to binarize the peak pixel based on the pixel value of each pixel in the evaluation pixel group (for example, each evaluation pixel group consisting of pixel P [x−2, y] to pixel P [x + 2, y]) Since the horizontal peak pixel P [x, y] is binarized based on the pixel value of the pixel), the change in shading near the peak pixel can be taken into consideration in detail, and blurring and squashing are unlikely to occur (in other words, , Good binarization can be realized.

  In the first embodiment, the evaluation pixel group is formed of five pixels including a peak pixel (horizontal or vertical peak pixel) as a central pixel. However, the evaluation pixel group is a peak pixel (horizontal or vertical peak pixel). ) As a central pixel.

  The predetermined direction in the binarization method W may be, for example, the horizontal direction or the vertical direction in the first embodiment, and may be considered to include the horizontal direction and the vertical direction. In the first embodiment, for example, when the pixel P [x, y] is extracted as a horizontal peak pixel which is a kind of peak pixel (step S11), the pixels P [x−2, y] to P [x + 2, y]. Based on the pixel values v [x−2, y] to v [x + 2, y] of each pixel in the evaluation pixel group including the horizontal peak pixel P through steps S12 and S13 and the processing after step S21 in FIG. [X, y] is binarized. The binarization process according to the first embodiment may be considered to include all or part of the processes of steps S25, S26, S28, and S30.

Specifically, for example, in the binarization method W, the evaluation pixel group includes a first pixel and a second pixel located on both sides of the peak pixel in the predetermined direction, and the peak pixel in the predetermined direction. A third pixel located outside the first pixel and a fourth pixel located outside the second pixel (for example, located farther from the first pixel when viewed from the peak pixel in the predetermined direction) Including a third pixel and a fourth pixel positioned farther than the second pixel; more specifically, for example, including a third pixel and a fourth pixel positioned one pixel apart from the peak pixel in the predetermined direction ), The first pixel is located between the peak pixel and the third pixel, and the second pixel is located between the peak pixel and the fourth pixel. In the binarization step, the peak image is Based on; first to fourth difference between the pixel value (see FIG. 7 (a) or (b) for _{example, DIF L1, DIF R1, DIF} L2, DIF R2) between the first to fourth pixel and The peak pixel may be binarized.

  Rather than considering only the first difference and the second difference, or only the third difference and the fourth difference, the peak pixel is binarized based on a total of four differences consisting of the difference of two pixels on both sides of the peak pixel. Therefore, it is possible to consider in detail the density change in the vicinity of the peak pixel, and it is possible to realize good binarization that is less likely to cause blurring or squashing (in other words, the blurring or squashing is eliminated).

In the first embodiment, for example, when the pixel P [x, y] is extracted as a horizontal peak pixel which is a kind of peak pixel (step S11), the pixel values v [x-2, y] to v [x + 2, y ] Based on the first to fourth differences (for example, DIF _L1 , DIF _R1 , DIF _L2 , DIF _R2 ; see FIG. 7A or FIG. 7B), Steps S12 and S13 and Step S21 in FIG. The horizontal peak pixel P [x, y] is binarized through the process (for example, steps S25 and S26).

More specifically, for example, the binarization method W includes a peak intensity derivation step (steps S12 and S15) for deriving the sum of the first edge intensity and the second edge intensity at the peak pixel as the peak intensity of the peak pixel. Further, in the peak intensity derivation step, the first difference (for example, DIF _L1 ) and the third difference (for example, DIF _L2 ) have the same sign or the first difference is zero and the third difference When the absolute value of the first difference is larger than the absolute value of the first difference, the third difference is derived as the first edge strength (for example, the left edge strength), while otherwise, the first difference is calculated as the first edge. Derived as an intensity, the second difference (for example, DIF _R1 ) and the fourth difference (for example, DIF _R2 ) have the same sign or the second difference is zero and the absolute value of the fourth difference When the value is larger than the absolute value of the second difference, the fourth difference is derived as the second edge strength (for example, right edge strength), while otherwise, the second difference is used as the second edge strength. Derived. At this time, in the binarization step, when the absolute value of the peak intensity is equal to or greater than a predetermined value, the peak pixel may be binarized based on the sign of the peak intensity.

  Thereby, it is possible to realize good binarization that is less likely to cause blurring or crushing even for an input image as described with reference to FIGS.

  In the first embodiment, for example, the pixel P [x, y] is extracted as a horizontal peak pixel which is a kind of peak pixel (step S11), and the horizontal peak intensity of the horizontal peak pixel P [x, y] is a predetermined value. At the above, the horizontal peak pixel P [x, y] is binarized through the classification of the horizontal peak pixel P [x, y] according to the sign of the horizontal peak intensity into the blurred candidate pixel or the collapse candidate pixel. (For example, steps S25 and S26).

  And, for example, in the binarization step according to the binarization method W, a peak pixel whose absolute value of the peak intensity is less than the predetermined value and pixels other than the peak pixel are binarized using a threshold value. (Step S34).

  In the binarization method W, the threshold value may be a preset fixed value, or may be set based on the input image.

  The embodiment of the present invention can be appropriately modified in various ways within the scope of the technical idea shown in the claims. The above embodiment is merely an example of the embodiment of the present invention, and the meaning of the term of the present invention or each constituent element is not limited to that described in the above embodiment. The specific numerical values shown in the above description are merely examples, and as a matter of course, they can be changed to various numerical values.

DESCRIPTION OF SYMBOLS 1 Character recognition apparatus 10 Input image acquisition part 20 Binary image generation part 21 Peak pixel extraction part 22 Peak intensity | strength derivation part 23 Binary part 30 Character recognition process part 40 Output part

Claims (7)

A binarization method for binarizing a multi-value input image,
A peak pixel extraction step of extracting, as a peak pixel, a pixel that takes an extreme value when the input image is scanned along a predetermined direction;
A binarization step of binarizing the peak pixel based on a pixel value of each pixel in an evaluation pixel group including five pixels continuously arranged along the predetermined direction and including the peak pixel as a central pixel And a binarization method comprising: The evaluation pixel group includes a first pixel and a second pixel located on both sides of the peak pixel in the predetermined direction, a third pixel located outside the first pixel when viewed from the peak pixel in the predetermined direction, and A fourth pixel located outside the second pixel, wherein the first pixel is located between the peak pixel and the third pixel, and the second pixel is located between the peak pixel and the fourth pixel. And
2. The binarization step binarizes the peak pixel based on first to fourth differences in pixel values between the peak pixel and the first to fourth pixels. The binarization method described in 1. A peak intensity deriving step of deriving the sum of the first edge intensity and the second edge intensity at the peak pixel as the peak intensity of the peak pixel;
In the peak intensity derivation step, the first difference and the third difference have the same sign or the first difference is zero and the absolute value of the third difference is greater than the absolute value of the first difference When deriving the third difference as the first edge strength, otherwise deriving the first difference as the first edge strength,
When the second difference and the fourth difference have the same sign or the second difference is zero and the absolute value of the fourth difference is larger than the absolute value of the second difference, the fourth difference is While deriving as the second edge strength, otherwise deriving the second difference as the second edge strength,
3. The binarization according to claim 2, wherein, in the binarization step, when the absolute value of the peak intensity is equal to or greater than a predetermined value, the peak pixel is binarized based on a sign of the peak intensity. Method. 4. The binarization step, wherein a peak pixel whose absolute value of the peak intensity is less than the predetermined value and pixels other than the peak pixel are binarized using a threshold value. The binarization method described. The binarization method according to claim 4, wherein the threshold value is a preset fixed value or is set based on the input image.   The program for making a computer implement | achieve the binarization method in any one of Claims 1-5. A binarization device for binarizing a multi-value input image,
An input image acquisition unit for acquiring the input image;
A peak pixel extraction unit that extracts, as a peak pixel, a pixel that takes an extreme value when the input image is scanned along a predetermined direction;
A binarization unit that binarizes the peak pixel based on a pixel value of each pixel in an evaluation pixel group that includes five pixels continuously arranged along the predetermined direction and includes the peak pixel as a central pixel And a binarization apparatus comprising: