JP2001160903A - Image correction method and device, and recording medium with program for executing its method recorded thereon - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an image correction method and device for eliminating highly-dense noises. SOLUTION: Smooting an original image which is a binary image converts the image once into a multi-gradation contrast image. This contrast image is divided into blocks, a mean gray scale for each block is calculated, a mean gray scale between the gray scale of a target block and the gray scale among blocks which are adjacent horizontal and vertical to the target block is obtained, the resulting gray scale is used for a threshold corresponding to all pixels in the target block, and a threshold image consisting of the threshold is generated. Finally, the magnitude correlation between each pixel value of the contrast image and a pixel value of the threshold image corresponding thereto is checked, '1' is set, when greater and '0' is set in other cases, the contrast image is restored again to the binary image. Thus, high dense noises that cannot be eliminated by a conventional technology, such as high dense small white dot noises of a graphic part and highly dense small black dot noise on a background part can be eliminated. As a result, when applying the image processor to a character recognition device or the like, errors in segmentation of characters of a received document image and errors in recognition of the characters from a scanner or a facsimile terminal can be decreased.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image correction method and apparatus for a binary image such as a document input by a scanner or a facsimile terminal and for removing noise from the image to improve the quality of the image. Things.

[0002]

2. Description of the Related Art As a conventional technique for removing noise from a binary image of a document input by a scanner or a facsimile and removing noise from the image, there is a method for removing isolated noise (supervised by Soken Publishing, Hideyuki Tamura, Japan Industrial Technology Center, “Introduction to Computer Image Processing,” p. 107). This method finds a black pixel surrounded by white pixels and replaces it with white pixels as noise, and finds a white pixel surrounded by black pixels and converts it to noise. This is a method of replacing black pixels. FIG. 5 shows an example of the process.

FIG. 5A shows an input original image. One square represents one pixel, and pixels surrounded by a dotted-line circle are noise. In the conventional method for this image, it is checked whether or not each black pixel is isolated from other black pixels. FIG.
In the image shown in FIG. 5A, as shown in FIG. 5B, all eight pixels around the noise pixel are white pixels and are isolated black pixels. By replacing pixels, noise can be removed as shown in FIG. Also, in the image of FIG. 5D, one white pixel in the square constituted by the central black pixel is noise, but for this image, eight white pixels around the noise pixel Since all of the pixels are black pixels, it is understood that this is noise, and by replacing this with black pixels, noise can be removed as shown in FIG.

[0004]

However, the above-mentioned conventional method has a problem that high-density noise cannot be removed. Here, the high-density noise refers to "white sesame noise in a black portion of a figure or the like" or "black sesame noise in a white portion of a background or the ground". In addition, high density means that the ratio of the number of pixels of noise per unit area is high, and is not necessarily noise having high spatial frequency. As an example,
FIG. 7 shows an image obtained by superimposing high density white pixel noise on the original image shown in FIG. In such an input image, since the noise is not isolated, the above conventional method has no effect. Specifically, such high-density noise causes a problem in a document recognition device that cuts out and recognizes a character written in a document image input from, for example, a facsimile or a scanner. In this case, it appears as high-density white sesame noise in the figure part and high-density black sesame noise in the ground part, and these are major causes of character cutout and recognition errors. .

[0005] An object of the present invention is to provide an image correction method and apparatus capable of removing high-density noise that cannot be removed by the above-mentioned conventional technology.

[0006]

SUMMARY OF THE INVENTION The above-mentioned object of the present invention is as follows.
The problem is solved by the means listed below.

One of the means is an image correction method for removing noise from an original image which is a binary image. The original image is converted into a multi-tone image by performing smoothing on the original image. A multi-gradation step, for each partial region of the grayscale image, determine a local threshold for the partial region based on the frequency of appearance of the density of pixels in the partial region, and, based on the determined local threshold, A threshold image creating step of creating a threshold image, which is an image composed of threshold values for each pixel of the gray image, and comparing each pixel of the gray image with a threshold value of a corresponding pixel of the threshold image, A binarizing step of creating a value image.

Alternatively, in an image correction apparatus for removing noise from an original image which is a binary image, a multi-tone image is converted into a multi-tone image by smoothing the original image. Means for determining a local threshold value for each partial area of the grayscale image based on the frequency of occurrence of the density of pixels in the partial area, and based on the determined local threshold value, A threshold image generating means for generating a threshold image, which is an image composed of thresholds for each pixel of the image; comparing each pixel of the grayscale image with a threshold of a corresponding pixel of the threshold image; An image correction apparatus comprising: a binarizing unit that creates the image.

Alternatively, in an image correction method for removing noise from an original image which is a binary image, a procedure for converting the original image into a multi-tone image by performing smoothing on the original image; For each partial region of the gray image, determine a local threshold for the partial region based on the appearance frequency of the density of pixels in the partial region, based on the determined local threshold,
Creating a threshold image, which is an image composed of threshold values for each pixel of the gray image, and comparing each pixel of the gray image with the threshold value of the corresponding pixel of the threshold image to create a binary image based on the magnitude And a program for executing the image correction method, wherein the program for executing the steps of the above-described steps is recorded on a computer-readable recording medium.

In the present invention, a binary image, which is an original image, is once converted into a grayscale image by smoothing, and determined based on the frequency of appearance of the density of pixels in a partial area (block) of the grayscale image. In a conventional technique, a threshold image composed of a threshold value for each pixel of a grayscale image is created based on the local threshold value, the magnitude of the corresponding pixel value of each image is checked, and the binary image is returned again. Eliminate high-density noise that cannot be eliminated. By its operation, when the present invention is applied to a document recognition method, a device or a storage-type facsimile exchange method and a device that cut out and recognize characters such as input form images from a scanner or a facsimile,
A high-density white sesame noise in a figure part and a high-density black sesame noise in a ground part are removed, thereby making it possible to reduce character cutout and recognition errors.

[0011]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a processing block diagram according to an embodiment of the present invention. In FIG. 1, reference numeral 1 denotes a multi-gradation unit, 2 denotes a threshold image generation unit, and 3 denotes a binarization unit. Hereinafter, a detailed operation of the processing block of FIG. 1 will be described.

FIG. 2 shows an example of the processing flow of the processing block of FIG. Hereinafter, description will be made with reference to FIG. First, it is assumed that a binary image to be subjected to noise removal has been input to the multi-gradation means 1. The image in FIG. 7 is used as an example. FIG. 3 shows an example of the operation of the multi-gradation generating means 1.

The multi-gradation means 1 converts a binary image input as an original image into a gray-scale image. For this purpose, in the present embodiment, a filter whose example is shown in FIG. 3A is used. In this case, three pixels vertically and three pixels horizontally, but any size can be used. Also, various combinations can be used as long as the numerical value in the filter is large at the center and small at the periphery. The multi-gradation means 1 performs the following processing on each pixel of the input image which is a binary image (this is referred to as a pixel k). First, a set of pixels having the same size as the filter is cut out around the pixel k to form an array. X in FIG. 3 (b)
The result of performing this processing on the pixel indicated by is shown in FIG.
(C). Next, the cut array is multiplied by the value of the corresponding element of the filter, and the results are all added. When this processing is performed between FIG. 3A and FIG. 3C, 0.1 × 0 + 0.5 × 0 + 0.5 × 0 + 2 × 0 + 0.5
× 0 + 0.1 × 0 + 0.5 × 0 + 0.1 × 1 = 0.1, which is 0.1. This is set as a new value corresponding to the original pixel k. This process is performed for all the pixels of the input image, and a new value corresponding to each pixel is collected to create an image, as shown in FIG. 3D (the numerical values represent shading). This is a grayscale image.

Next, the process proceeds to the processing of the threshold image creating means 2.
The threshold image creating means 2 determines a local threshold value for each partial area (block) of the grayscale image based on the appearance frequency of the pixel density in the partial area, and determines the determined local threshold value. , A threshold image, which is an image composed of threshold values for each pixel of the grayscale image, is created. Various concrete processing contents of the threshold image creating means 2 are conceivable. For example, a method shown in Japanese Patent Publication No. 14374/1994, "Binarization method of gray image" is also an example.

In one embodiment of the threshold image creating means 2,
The following operation processing is performed. The process is shown in FIG.
(A) to (d). The “frequency of appearance of pixel density” in the present embodiment is a histogram with luminance on the horizontal axis. This histogram generally has a distribution having two or more peaks for a white portion such as a background and a black portion such as a character.
A valley between the two or more peaks is found, and the luminance of the valley is set as a threshold. The simplest method is a method of calculating an average value as follows. In the present embodiment, the “local threshold” is “the average density itself for each divided block”. Furthermore, from "local threshold" to "threshold image"
Is obtained by smoothing the average value of the upper, lower, left, and right blocks.

First, the grayscale image shown in FIG. 3D (the squares indicate pixels, and the numerical values in the squares indicate shades) is divided into a plurality of blocks. FIG. 4A shows an example of such a case, which is divided into 3 × 3 blocks divided by a thick line. Next, an average value of density is calculated in each block. This example is shown in FIG. Next, with respect to the data for which the average value of the density was obtained in each block, for each average value, the average of the average values of the upper, lower, left, and right blocks and the average with itself were obtained, and this was calculated for all the A threshold image is created as a threshold corresponding to a pixel. For example, in the upper right block (value of 0.0) in FIG. 4B, there are only left (0.6) and lower (0.2) left, right, top, and bottom blocks, Average value (0.6 + 0.2) / 2 =
0.4 is obtained first, and then the average value of this 0.4 and its value 0.0 is (0.4 + 0.0) /2=0.2. FIG. 4 (c) shows the result corresponding to each block, and FIG. 4 (d) shows the threshold image created thereby.

As another embodiment of the threshold image creating means 2, Japanese Patent Publication No. 6-14374, "Binarization method of gray image"
An example in which processing is performed using the method indicated by.

First, a grayscale image is divided into blocks of a predetermined size. This division processing is not necessarily performed in FIG.
The regions need not be equally divided as shown in (a), and the regions may overlap each other or may be separated from each other.
Next, a threshold value determination process is performed. In this process, a threshold value is determined for each divided block from the frequency of appearance of gray values of pixels in the block. Various processes can be considered as a process for determining the threshold value from the appearance frequency of the density value. For example, as described above, the average value of the density values, or the median value, the average value of the maximum value and the minimum value, and the like are set as the threshold values. A simple method or a two-class problem in which threshold processing is separated into two classes, black and white, has the best degree of separation (maximum variance between classes)
There is a method using a density value as a threshold value (discrimination analysis method). Next, a weight value determination process is performed. In this processing, for each divided block, the average value of the brightness of the pixels in the block and the variance thereof are obtained, and the weight value indicating the contrast is determined based on this. Types of weight values include those using variance and standard deviation and those obtained by dividing (normalized) the variance and standard deviation by the average value of brightness. Next, a first interpolation process is performed. This process is based on the threshold value obtained in the threshold value determination process and the weight value obtained in the weight value determination process.
This is a process for correcting the threshold. By repeating this process, the threshold value of the block to which a large weight value has been given is propagated to the surrounding blocks. For example, the threshold value of a small area that does not include characters is corrected, and background noise can be removed. Next, a second interpolation process is performed. In this processing, it is assumed that the threshold value corrected by the first interpolation processing is given to the center of any four adjacent blocks, and a square having the center of the four blocks as a vertex is considered. A threshold value corresponding to an arbitrary pixel is obtained by interpolating the four threshold values at the vertices, and a threshold image is created as a threshold value corresponding to all pixels in the rectangle.

Next, the processing proceeds to the binarizing means 3. Here, the values of the corresponding pixels of the grayscale image and the threshold image are compared, and if the density of the pixel of the grayscale image is larger than the threshold of the corresponding pixel of the threshold image, the density is set to 1; Create a value image. 3A and FIG.
FIG. 4E shows this process performed using the threshold image of FIG. 4D, where density 1 is represented by black pixels and density 0 is represented by white pixels.
FIG. 4D shows that high-density white noise has been removed, and the effect of the present invention is clear.

It goes without saying that some or all of the functions of each unit shown in FIG. 1 can be realized by using a computer, or that the processing procedure shown in FIG. 2 can be executed by a computer. In addition, a program for realizing the function of each part by the computer or a program for executing the processing procedure by the computer is stored in a storage medium readable by the computer, for example, FD (Floppy Disk), MO, ROM, memory card, CD, DVD,
It can be recorded on a removable disk, provided, and distributed.

[0022]

As described above, according to the present invention, high-density noise that cannot be eliminated by the prior art, that is, high-density white sesame noise in a figure portion or high-density noise in a ground portion, Black sesame noise can be removed. As a result, a document recognition method and apparatus that recognizes by cutting out characters written on a form image input from a facsimile or scanner, or a storage-type facsimile exchange method that recognizes a document transmitted from a facsimile terminal, or the document If the method according to the present invention is applied to a storage-type facsimile exchange device having a recognition function, or if the device according to the present invention is incorporated in those devices, high-density white sesame noise or the like in a figure part can be obtained. In addition, high-density black sesame noise in the ground portion is removed, thereby reducing character segmentation and recognition errors, and improving performance.

[Brief description of the drawings]

FIG. 1 is a processing block diagram showing an embodiment of the present invention.

FIG. 2 is a diagram showing an example of a processing flow according to an embodiment of the present invention.

FIGS. 3 (a), (b), (c), and (d) are diagrams showing processing steps of a gradation conversion unit according to the present invention.

FIG. 4 shows (a), (b), (c), (d) and (e)
FIG. 4 is a diagram illustrating a process of a threshold image creating unit and a binarizing unit according to the present invention.

FIG. 5 shows (a), (b), (c), (d), and (e)
It is a figure showing the processing process of the conventional method.

FIG. 6 is a diagram illustrating an example of an input image.

FIG. 7 is a diagram illustrating an example in which high-density noise is superimposed on an input image.

[Explanation of symbols]

 DESCRIPTION OF REFERENCE NUMERALS 1: gradation conversion means 2: threshold image creation means 3: binarization means

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Minoru Mori, 2-3-1, Otemachi, Chiyoda-ku, Tokyo Nippon Telegraph and Telephone Corporation (72) Inventor Tetsuya Kibuchi 2--3, Otemachi, Chiyoda-ku, Tokyo No. 1 Within Nippon Telegraph and Telephone Corporation (72) Inventor Sakuichi Otsuka 2-3-1 Otemachi, Chiyoda-ku, Tokyo F-Term within Nippon Telegraph and Telephone Corporation (reference) 5B057 AA11 BA11 BA24 BA30 CA02 CA06 CA12 CA16 CB02 CB06 CB12 CB16 CC02 CE05 CE12 CH01 CH11 DB02 DB05 DB09 DC23 5C077 LL05 LL06 MP04 NP01 PP02 PP21 PP46 PP68 PQ12 PQ19 PQ22 RR02 RR07 RR11 RR15

Claims (3)

[Claims] 1. An image correction method for removing noise from an original image which is a binary image, wherein the original image is converted into a multi-tone image by performing smoothing on the original image. And determining a local threshold for the partial area for each partial area of the grayscale image based on the frequency of appearance of the density of pixels in the partial area, and based on the determined local threshold, A threshold image creating step of creating a threshold image that is an image composed of thresholds for each pixel of the image; comparing each pixel of the grayscale image with a threshold value of a corresponding pixel of the threshold image; Create 2
An image correction method, comprising: 2. An image correction apparatus for removing noise from an original image which is a binary image, wherein the original image is converted to a multi-tone image by smoothing the original image. Means for determining a local threshold value for each partial region of the grayscale image based on the frequency of appearance of the density of pixels in the partial region, and based on the determined local threshold value, Threshold image creating means for creating a threshold image, which is an image composed of threshold values for each pixel of the image; comparing each pixel of the grayscale image with a threshold value of a corresponding pixel of the threshold image; Create 2
An image correction apparatus comprising: a value conversion unit. 3. An image correction method for removing noise from an original image that is a binary image, wherein the original image is converted into a multi-tone image by performing smoothing on the original image. For each partial area of the grayscale image, determine a local threshold value for the partial area based on the frequency of appearance of the density of pixels in the partial area, and, based on the determined local threshold, each pixel of the grayscale image A step of creating a threshold image, which is an image composed of a threshold value, and a step of comparing each pixel of the grayscale image with a threshold value of a corresponding pixel of the threshold image, and creating a binary image according to the magnitude of the computer. A recording medium on which a program for executing an image correction method is recorded, wherein the program to be executed by the computer is recorded on a recording medium readable by the computer.