JPH04278688A - Adaptive binary system for gradation picture - Google Patents

Abstract

PURPOSE:To damp only the noise of an entry frame, etc., to restore the break and blur of a character and to prevent the burden from being given to the recognition processing at the rear step. CONSTITUTION:An original image 1 having the blur and break of a character and the noise such as the net point-shaped pattern, when it is processed by the neighbourhood concentration comparing processing, becomes a binary image 2, and sent to the neighbourhood area searching processing. A binary image 3 processed by the dot extracting processing and a binary image 4 processed by the neighbourhood concentration large bias comparing processing are both sent to the noiseless image forming processing, and a binary image 6 formed by the processing is sent to the neighbourhood area searching processing. A binary image 7 processed by the neighbourhood area searching processing is sent to the except-inside preventing processing. Together with this, a binary image 5 formed by the except-inside preventing image forming processing is also sent to the except-inside preventing processing and the image processed here becomes a binary processing image 8.

Description

[Detailed description of the invention]

0001

[Industrial Application Field] The present invention reads handwritten characters, etc. on various forms, etc., including the entry frames, as a multivalued grayscale image, and performs multivalued gray scale images as a preprocessing for recognizing handwritten characters, etc. The present invention relates to a method for converting a grayscale image into a binary image.

0002

[Prior Art] For example, there is a method disclosed in Japanese Patent Application Laid-Open No. 1-116782 as a method for recognizing monetary figures handwritten on checks in the United States. It is necessary to convert the multi-value gray image into binary data as pre-processing. In this binarization, a fixed threshold value is determined from the gradation distribution state of the read density data, and the binarization is performed based on the magnitude of the threshold value.

0003

[Problems to be Solved by the Invention] However, since the characters are handwritten, there may be blurred or broken characters in some cases, and these parts are erased by binarization. Additionally, characters that are likely to be interrupted can be corrected by providing a separate pattern in the recognition process, but this is not perfect against blurred characters, resulting in erroneous recognition or unrecognizability.

[0004]Furthermore, in addition to thin lines, the text frame has a halftone dot pattern, and many checks have a halftone dot pattern on the background (ground), and the background pattern is not uniform. do not have. For this reason, these dot-shaped parts are 2
They often remain as noise without disappearing when converted into values, placing a burden on subsequent recognition processing, and often resulting in erroneous recognition or unrecognizability.

The present invention has been made in view of the above-mentioned problems, and an object of the present invention is to eliminate only the noise in the entry frame, etc., and to repair the broken and blurred characters, thereby eliminating the burden on the subsequent recognition processing. The purpose of the present invention is to provide an adaptive binarization method.

[0006]

[Means for Solving the Problems] The present invention provides thin lines, broken lines,
A writing frame is designated by halftone dots, etc., and a plurality of characters freely handwritten on a background having a halftone pattern are read and binarized as a multivalued gray image including the entire writing frame. The above-mentioned object of the present invention is to compare each pixel of the gray scale image with the density of the vicinity of the pixel, determine a threshold value for each pixel, and binarize the pixel. Even if noise removal is incomplete, a first feature image is created in which the blurred parts of the characters are repaired, and a binary image is created in which both the blurred parts of the characters and the noise are removed from each pixel of the grayscale image. By creating a second characteristic image of This is achieved by allowing the image to be obtained.

[0007]

[Operation] In the present invention, to binarize a grayscale image, instead of using a fixed or variable threshold value, an adaptive method is used that takes into consideration the situation around the pixel. In other words, two types of neighboring density comparison processing with different bias values are performed on the original image (shaded image), and dot extraction processing is performed in parallel, followed by noiseless image creation processing and neighboring area search processing. ing. In addition, a hollow-hole prevention image is created to remove the hollow areas that occur in the thicker parts of the image, in order to complete the image.

[0008]

[Example] Fig. 1 is a flow chart showing an overview of the binarization method of the present invention. The gradation image is stored for each pixel in 64 levels of gradation, with the darkest part being 0 and the brightest part being 63, and this original image (gradation image) 1 is processed through neighboring density comparison processing (step S10), The dot extraction process (step S20), the neighboring density large bias comparison process (step S30), and the void prevention image creation process (step S40) are processed in parallel. When processed in the neighborhood density comparison process (step S10), 2
The value image becomes value image 2, and its data is sent to the nearby area search process (step S60).

[0009] Also, dot extraction processing (step S20)
Binary image 3 processed in step 3 and binary image 4 processed in neighborhood density large bias comparison processing (step S30) are both sent to noiseless image creation processing (step S50), and the Value image 6 is sent to nearby area search processing (step S60). The binary image 7 processed in the neighborhood area search process (step S60) is sent to the blank-out prevention process (step S70). At the same time, the binary image 5 created in the image creation process (step S40) to prevent hollow areas is also processed (step S7).
0) and the image processed here becomes a binary processed image 8.

FIG. 2 shows the neighboring density comparison process (step S10).
) is a flowchart showing an example of the operation. First, a pixel number is set to i (step S11), and (D) of FIG.
As shown in (E) and (C) and (D) of FIG.
The maximum value of pixel shading in a 3×3 area centered on (
The value of the brightest pixel in the area is set to max1 (step S12). Note that the shaded portions in (D) and (E) of FIG. 3 and (C) and (D) of FIG. 4 indicate images. Next, the maximum value of the shading of pixels in a 9×9 area centered on pixel ai is set to max4. In the case of a halftone pattern as shown in FIG. 3(D), the maximum values max1 and max
4 is as shown in FIG. 3(A), and the maximum values max1 and max4 in the case of a character part as shown in FIG. 3(E) are as shown in FIG.
(B). The broken line in FIG. 2B shows the case where the characters are blurred. Also, (C) in Figure 3
indicates the maximum values max1 and max4 when reading the background (earth).

As described above, the maximum values max1 and ma
After finding x4,

0012

[Equation 1] It is determined whether max1>(max4-bias) (step S14), and if so, it is assumed that it is not a character and the pixel ai is set to 0 (step S1
5) If the formula 1 does not hold, it is considered a character and the pixel ai
=1 (step S16). Therefore, (A
), (D) and (C), ai=0, and in cases (B) and (E) of FIG. 3, ai=1. In other words, since ai=1 for the faded part of the character, it is said that it has been repaired. By repeating the above operation for all pixels (steps S17 and S18), the entire image can be binarized, and no small halftone dot pattern remains as shown in image 2 in FIG.

However, in the case of the size relationship of a thick character part as shown in (C) of FIG. 4, the reading shading is as shown in (A) of FIG. 4, so Equation 1 holds and pixel ai=0 ,
Hollow holes (white areas in the image) as shown in Image 2 occur. Further, in the case of a large halftone pattern as shown in FIG. 4(D), the reading gradation is as shown in FIG. 4(B), so the equation 1 does not hold and the pixel ai=1. In other words, the large halftone pattern remains as noise. As described above, when the neighboring density comparison process (step S10) is applied to the original image 1, a binary image 2 is obtained. This binary image 2 corresponds to the first characteristic image. Although noise removal is incomplete in this neighborhood density comparison process, faded portions of characters are repaired.

FIG. 5 shows dot extraction processing (step S20)
6 is a flowchart showing an example of the operation. First, a pixel number is set to i (step S21), and then for 8 pixels aij (j=1 to 8) around pixel ai as shown in FIG.

0
015]

[Math 2] Calculate and

[0016]

[Math 3] Find a number N (integer) that satisfies the requirement (step S22).

[0017] And,

[Equation 4] Determine whether Δ≦−K1 (step S23), and if so, leave the pixel as pixel ai=1 (step S2
4). In other words, subtracting the gradations of surrounding pixels from pixel ai and finding that the sum is less than a certain value - K1 means that there are some pixels around ai that are brighter than ai, and in this case It is assumed that ai is a dot, and ai=1. The same applies to the pixel ai shown in FIG. If equation 4 does not hold in step S23,

001
8]

[Equation 5] Determine whether N≧K3 (step S26), and if so, proceed to step S24, otherwise the pixel a
Erase the pixel by setting i=0 (step S27)
. In other words, even if the sum is not less than -K1, the pixel ai
Pixels much brighter than (

If there is a constant value K3 or more around ai (which also satisfies Equation 3), ai is regarded as a dot and ai=1. After the pixel ai is left as 1 in step S24, aij (j=1 to 8) is set to 1 so that surrounding pixels are also left (step S25). This is to represent the dots more reliably. The above operation is repeated until all pixels are completed (steps S28, S29).

Dot extraction processing as described above (step S
20) Dots as shown in the shaded area in FIG. 6 and thin lines as shown in the shaded area in FIG. 7 are extracted. Therefore, the original image 1 in FIG. 1 becomes a binary image 3 in the dot extraction process (step S20).

Neighboring concentration large bias comparison process (step S
30) is the above-mentioned neighborhood density comparison process (step S10).
), the value of bias in Equation 1 in step S14 becomes relatively large, and bias2(>
The only difference is that it is (bias). Figure 8 explains the operation, and the slightly larger dots as shown in Figure (C) are also similar to those shown in Figure (A).

0021
]

[Equation 6] Since max1>(max4-bias2), the pixel ai is erased as 0. This is the bias size of the neighboring density comparison process (step S10).

[0022]

[Equation 7] Since max1<(max4-bias), the pixel ai=1 remains. Furthermore, since Equation 5 also holds true in the case of the faded character shown in FIG. 8(D), the pixel ai is erased as 0. FIG. 8B shows the same faded characters as the broken line in FIG. 3B.

As described above, in the neighboring density large bias comparison process (step S30), due to the large value of bias2, slightly large dots and blurred characters disappear. Therefore, the original image 1 in FIG.
becomes.

Binary image 3 created in the dot extraction process (step S20) and neighboring density large bias comparison process (
The binary image 4 created in step S30) is processed in a noiseless image creation process (step S50) to create a noise-free binary image 6. This binary image 6 corresponds to the second characteristic image. In addition to noise, blurred parts of characters are also removed. This situation will be explained with reference to the flowchart in FIG.

First, the “1” pixel of image 4 and the “1” pixel of image 3
” pixel by pixel (step S51), and determines whether the pixel of image 3 corresponding to the pixel of “1” of image 4 is “1” (step S52).If the corresponding pixel is “1”
”, the pixel “1” of image 4 is erased and set to “0” (step S53), and if the corresponding pixel is not “1”, the pixel “1” of image 4 is left as is (step S53).
4). The operations of steps S52 to S54 described above are performed on image 4.
This process is repeated until all "1" pixels have been dealt with (step S55).

Noiseless image creation processing (step S50)
The binary image 6 created in ) is processed in the neighborhood area search process (step S60) together with the binary image 2 created in the neighborhood density comparison process (step S10). That is, as shown in the flowchart of FIG.
and image 6 (step S61), and “
A 7×7 area is searched for the pixel of image 6 corresponding to the pixel of “1” (step S62). FIG. 11 shows this situation. Next, if there is one pixel of “1” in the above area, It is determined whether or not there are more than 1 (step S63), and if so, the "1" pixel of image 2 is left as is (step S64), and if not, the "1" pixel of image 2 is deleted and "0" (step S65). For example, pixel A indicated by an , On the other hand, pixel B indicated by the X mark has no "1" pixels in the corresponding 7x7 area, so pixel B is "1".
is erased and set to “0”. Above steps S62 to S65
This is repeated until all "1" pixels of image 2 have been dealt with (step S66). As a result, noise is removed, breaks due to blurring etc. are interpolated, and an image as shown in FIG. 1 is obtained.

On the other hand, in FIGS. 4(A) and 4(C), it has been explained that in the neighboring density comparison process (step S10), a hollow part occurs in the thick part, but in order to solve this problem, the present invention A dropout prevention process (step S70) is performed.

As a premise of such processing, in the present invention, the original image 1 is subjected to hollow-hole prevention image creation processing (step S40).
). That is, as shown in the flowchart of FIG. 12, a pixel number is first set to i (step S41).
, the average value am of the shading of pixels within a 5×5 area centered on pixel ai is calculated (step S42). Figure 13
shows the situation. Then, it is determined whether the average value am is smaller than the threshold value K4 (step S43).
If so, the pixel ai is left unchanged as 1 (step S44), and if the average value am is equal to or higher than the threshold value K4, the pixel ai is erased and set to "0" (step S45). By repeating the above steps S42 to S45 for all pixels (steps S46 and S47), it is possible to create an image 5 to prevent hollow spots as shown in FIG.

The hollow-hole prevention image 5 created as described above is subjected to a hollow-hole prevention process (step S) together with the binary image 7.
70) and are added to obtain the processed image 8 as shown in FIG.

[0030]

As described above, in the present invention, neighboring density comparison processing, dot extraction processing, and neighboring density large bias comparison processing are performed in parallel on a grayscale image, and furthermore, noiseless image creation processing and neighboring area search processing are performed in parallel. Since a binary processed image is obtained by performing this process, only characters can be reliably binarized even if noise, blurring, etc. are present. In addition, hollow portions occur in thick character portions, but hollow portions can be prevented by creating and adding a hollow portion prevention image.

[Brief explanation of the drawing]

FIG. 1 is a flowchart schematically showing a binarization method of the present invention.

FIG. 2 is a flowchart illustrating an operation example of neighboring density comparison processing.

FIG. 3 is a diagram for explaining the operation of neighboring density comparison processing.

FIG. 4 is a diagram for explaining a problem in neighboring density comparison processing.

FIG. 5 is a flowchart showing an example of the operation of dot extraction processing.

FIG. 6 is a diagram for explaining dot extraction processing.

FIG. 7 is a diagram for explaining dot extraction processing.

FIG. 8 is a diagram for explaining nearby concentration large bias comparison processing.

FIG. 9 is a flowchart showing an operation example of noiseless image creation processing.

FIG. 10 is a flowchart illustrating an operation example of nearby area search processing.

FIG. 11 is a diagram for explaining nearby area search processing.

FIG. 12 is a flowchart illustrating an example of the operation of image creation processing to prevent hollow defects.

FIG. 13 is a diagram for explaining image creation processing to prevent hollow portions.

FIG. 14 is a flowchart showing blank omission prevention processing.

[Explanation of symbols]

1 Original image (shaded image) 2 Binary image 3 Binary image 4 Binary image 5. Hollow-out prevention image 8 Processed image (binary image)

Claims (1)

[Claims] Claim 1: A method in which a writing frame is designated and a plurality of characters handwritten on a background having a halftone pattern are read as a multi-valued gray image including the entire writing frame and then binarized. In this step, for each pixel of the grayscale image, a threshold value is determined for each pixel by comparing it with the density in the vicinity of that pixel, and the noise is binarized. creating a first feature image in which the blurred portions of the characters are repaired, and creating a binary second feature image in which both the blurred portions of the characters and the noise are removed from each pixel of the grayscale image; By combining the first feature image and the second feature image, only the noise is removed and the blurred portions of the characters are repaired to obtain an optimal binary image. An adaptive binarization method for grayscale images.