CN106127817A - A kind of image binaryzation method based on passage - Google Patents

Abstract

The invention discloses a channel-based image binarization method, comprising steps: for an image that needs to be binarized, the image is regarded as a pixel matrix I, the width of the image is recorded as w, and the height is recorded as h; Each row processes the pixel matrix I, scans the pixels of one row each time, and collects the red, blue and green channel values and gray value information of the statistical pixel I[i][j] when processing the i-th row, and Calculate the average value of these four quantities in the i-th row of pixels; count the pixel values of each channel in the i-th row according to the average value and threshold of the corresponding channel for each channel of the i-th row of pixels; if i<h-1, i=i+1; then repeat step S2-S3 to perform channel-based binarization processing on the next row of pixels. Compared with the prior art, the invention can quickly and accurately binarize images with uneven illumination.

Description

A channel-based image binarization method

technical field

The invention relates to the technical field of digital image processing, in particular to a channel-based image binarization method.

Background technique

Since the advent of computers and the introduction of artificial intelligence technology in the 1940s, computers have been expected to replace human brain work. Among them, image and graphics processing has been practiced and researched in computers. People hope that computers can automatically process images, and finally Present the results of the processing in front of people. Since the OCR technology was proposed in the 1920s to the present, OCR technology has reached a relatively high level, and a number of OCR products have also appeared, the most extensive of which is image license plate recognition, that is, to take pictures of the license plate number, and then pass Image processing technology for license plate number extraction. It can be seen that image processing technology occupies a certain position in the computer field. In image processing technology, image binarization is an important step in image preprocessing, and the quality of binarization directly determines the difficulty and accuracy of subsequent steps.

Although the current OCR technology research has been quite effective, it is still in the development stage, and technologies such as image preprocessing still need to continue research and practice. A good image binarization processing method can greatly improve the efficiency and accuracy of character feature extraction and recognition in images. Since IBM proposed the Chinese character recognition scheme in the 1960s, Chinese character recognition has been studied. In Chinese character recognition, each effective feature of Chinese characters needs to be extracted and matched with the Chinese characters in the feature library. In OCR technology, since the color image has a large amount of information, and in the feature extraction operation, the structural features of the character itself are needed, so binarizing the image is beneficial to reduce the complexity of subsequent operations. Binarization image processing is to analyze the pixel value of the input color image, and divide the foreground and background according to the threshold value obtained by a specific algorithm. The binarization process sets the text area of the image, that is, the foreground part, to black, and sets the background to white to highlight the text area, which greatly facilitates feature extraction and character recognition.

Binarization processing methods are mainly divided into two categories: global processing and local processing. In global processing, a threshold is used to judge the entire image. After the threshold is calculated, the threshold will remain unchanged until the image processing is completed. Among them are the Kittler algorithm and the histogram-based global threshold algorithm. Partial processing is dynamic threshold evaluation, which divides the image into roughly equal parts, calculates the threshold value of the current part each time, and then performs binarization processing until the entire image processing is completed. Among them are the Wall algorithm and the Wellner adaptive filtering threshold algorithm. In image processing, there are many factors that affect the quality of image processing. Different binarization algorithms are suitable for different situations. Although the above algorithm can solve most of the problems, as the complexity of the image increases, the processing results are still somewhat unsatisfactory, especially in the case of uneven illumination of the image, the resulting image after processing is blurred or mainly Parts are not highlighted etc. In the image processing of character recognition, the quality of the image directly affects the processing quality and recognition rate. Most of the time when taking pictures, we can't guarantee good image quality, the most common one is uneven lighting.

Contents of the invention

In order to overcome the deficiencies of the prior art and solve the image processing problem of poor binarization effect due to uneven illumination, the present invention proposes a channel-based image binarization method.

The technical scheme of the present invention is such: a kind of channel-based image binarization method comprises the following steps:

S1: For an image that needs to be binarized, the image is regarded as a pixel matrix I, the width of the image is recorded as w, and the height is recorded as h;

S2: Process the pixel matrix I line by line, scan the pixels of one line at a time, and collect the red, blue, green channel values and gray value of the statistical pixel point I[i][j] when processing the i-th line Information, respectively recorded as R _ij , G _ij , B _ij and GREY _ij , where GREY _ij is the gray value, GREY _ij = (R _ij +G _ij +B _ij )/3, and calculate this The average of the four quantities, namely:

redAvg _i = (∑ 0≤j _<w R _ij )/w,

greenAvg _i = (∑ 0≤j _<w G _ij )/w,

blueAvg _i = (∑ 0≤j _<w Bij)/w,

greyAvg _i = (∑ 0≤j _<w GREY _ij )/w;

S3: Count the pixel values of each channel in the i-th row according to the average value and threshold of the corresponding channel for each channel of the pixel point in the i-th row and perform binarization processing:

S31: record the sum of the red, green and blue channel values and the sum of the gray value of the pixel as redSum _i , greenSum _i , blueSum _i , greySum _i respectively; The number of pixels is respectively recorded as redCount _i , greenCount _i , blueCount _i , grayCount _i ; all of them are initialized to 0;

S32: Process the red channel R _i , use redAvg _i + α as the boundary value, traverse the i-th row of pixels, if R _ij <redAvg _i + α, then consider I[i][j] to be a red pixel, and set Add the red channel value to the sum redSum _i , add 1 to redCount _i ; use the same method to process G _i , B _i and GREY _i to get greenSum _i , greenCount _i , blueSum _i , blueCount _i , graySum _i , grayCount _i , where α is one-tenth of the standard deviation of the image;

S33: If redCount _i is 0, set the i-th row of pixels as white; if redCount _i is not 0, make redAvg _i equal to the average value of all red pixels, i.e. redAvg _i =redSum _i /redCount _i ; traverse the entire row pixels, if R _ij <redAvg _i +β, then set I[i][j] to black; where β is one-tenth of the standard deviation of the image;

S34: replace the red channel R _i with G _ij , B _ij , and GREY _ij channel information respectively, repeat steps S32-S33, and correct the binarization result. Unlike the red channel, only the original white pixel value is corrected to black;

S4: If i<h-1, i=i+1; repeat steps S2-S3, and perform channel-based binarization processing on the next row of pixels.

Further, step S2 is to scan the pixel matrix line by line from top to bottom.

Further, step S32 is to traverse the i-th row of pixels from left to right.

The beneficial effect of the present invention is that, compared with the prior art, the present invention can quickly and accurately binarize images with uneven illumination.

Description of drawings

FIG. 1 is a flowchart of the channel-based image binarization method of the present invention.

FIG. 2 is a detailed flow chart of step S3 in FIG. 1 .

Figure 3 is the receipt image before binarization.

Fig. 4 is the binarization result of the receipt picture in Fig. 3 after being processed by the method of the present invention.

detailed description

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

Referring to Fig. 1 and Fig. 2, a kind of channel-based image binarization method of the present invention comprises the following steps:

For an image that needs to be binarized, the image is regarded as a pixel matrix I, the width of the image is recorded as w, and the height is recorded as h. Each pixel, that is, each element I[i][j] in the matrix, where i∈[0,h), j∈[0,w), contains the information of the three channels of r, g, and b.

The pixel matrix I is processed row by row from top to bottom, scanning one row of pixels each time. When processing the i-th row (i∈[0,h)), collect statistics on the red, blue and green channel values and gray value information of I[i][j], which are recorded as R _ij , G _ij , B _ij and GREY _ij , where GREY _ij is the gray value, here adopt GREY _ij =(R _ij +G _ij +B _ij )/3 to calculate the average value of these four quantities in the i-th row of pixels, namely:

redAvg _i = (∑ 0≤j _<w R _ij )/w,

greenAvg _i = (∑ 0≤j _<w G _ij )/w,

blueAvg _i = (∑ 0≤j _<w Bij)/w,

greyAvg _i =(∑ 0 _≦j<w GREY _ij )/w.

Process and count the pixel values of each channel in the i-th row according to the average value and threshold of the corresponding channel for each channel of the i-th row of pixels:

Record the sum of the red, green, and blue channel values and the sum of the gray value of the pixel as redSum _i , greenSum _i , blueSum _i , and greySum _i ; pixels with red, green, blue, and gray values that meet the requirements The numbers are respectively recorded as redCount _i , greenCount _i , blueCount _i , grayCount _i ; all of them are initialized to 0.

Process the red channel R _i , use redAvg _i + α as the boundary value, traverse the i-th row of pixels from left to right, if R _ij <redAvg _i + α, then consider I[i][j] to be a red pixel , add the red channel value to the sum redSum _i , add 1 to redCount _i ; use the same method to process G _i , B _i and GREY _i to get greenSum _i , greenCount _i , blueSum _i , blueCount _i , graySum _i , grayCount _i , where α is the value with the best demarcation effect obtained from experiments, which is generally taken as one-tenth of the standard deviation of the image.

Correct the binarization result of I[i][j] according to the calculation results of the above steps:

If redCount _i is 0, all the i-th row pixels are set to white (255); if redCount _i is not 0, make redAvg _i equal to the average value of all red pixels, i.e. redAvg _i =redSum _i /redCount _i ; traverse the entire If R _ij <redAvg _i +β, then set I[i][j] to be black (0); where β is the value with the best adjustment effect obtained from the experiment, generally, β is ten times the standard deviation of the image one-third.

Then continue to modify the binarization results according to the G _ij , B _ij , and GREY _ij channel information. Unlike the r channel, only the original white pixels are corrected to black.

If i<h-1, i=i+1; continue to perform channel-based binarization on the next row of pixels.

In the following, the process of the present invention will be further elaborated by taking the supermarket receipt image as an example.

Figure 3 is an unprocessed supermarket receipt with uneven printing (illumination). It can be seen that the image of the receipt is uneven in brightness, and its brightness gradually decreases as the position goes down. The experimental value α for processing the red channel is -20, the experimental value α for the green channel is -25, the experimental value α for the blue channel is -23, and the experimental value α for the gray value is obtained from the experiment. The value is -20. Adjust and set the experimental value β of the black and white pixel values as follows: adjust the experimental value β of the red channel to 18, adjust the experimental value β of the green channel to 9, adjust the experimental value β of the blue channel to 8, adjust The grayscale experimental value β is set to 8.

Step 2, consider Fig. 3 as a pixel matrix I, the width of the image is recorded as w=350, and the height is recorded as h=500. Each pixel, that is, each element I[i][j] in the matrix, where i∈[0,500), j∈[0,350), contains the information of the three channels of r, g, and b.

Step 3, process the pixel matrix I line by line from 0 to 500, and scan the pixels of one line at a time. When processing the i-th row (i∈[0,500)), collect statistics on the red, blue and green channels of I[i][j] and the gray value information, which are recorded as R _ij , G _ij , B _ij and GREY _ij (Note: GREY _ij is the gray value, here adopt GREY _ij = (R _ij +G _ij +B _ij )/3), calculate the average value of these four quantities in the i-th row of pixels, namely:

redAvg _i = (∑ 0≤j _<350 R _ij )/350,

greenAvg _i = (∑ 0≤j _<350 G _ij )/350,

blueAvg _i = (∑ 0≤j _<350 B _ij )/350,

greyAvg _i =(∑ 0 _≦j<350 GREY _ij )/350.

Each channel of the i-th row of pixels is processed and counted according to the average value and threshold value of the corresponding channel:

1) Record the sum of the red, green, and blue channel values and the sum of the gray value of the pixel as redSum _i , greenSum _i , blueSum _i , and greySum _i respectively; red, green, blue, and gray values that meet the requirements The number of pixels is respectively recorded as redCount _i , greenCount _i , blueCount _i , grayCount _i ; all of them are initialized to 0.

2) Process the red channel R _i , use redAvg _i -20 as the boundary value, traverse the i-th row of pixels from left to right, if R _ij <redAvg _i -20, then consider I[i][j] to be a red pixel point, add the red channel value to the sum redSum _i , and add 1 to redCount _i ; use the same method to process G _i , B _i and GREY _i to get greenSum _i , greenCount _i , blueSum _i , blueCount _i , graySum _i , , greyCount _i .

For example, when i=200, when counting the 200th row, the experimentally obtained several average values are redAvg ₂₀₀ =138, greenAvg ₂₀₀ =168, blueAvg ₂₀₀ =148, and grayAvg ₂₀₀ =151.

Next, collect and count various pixel points such as red, green, blue, and gray in the 200th line, that is, they meet the conditions: R _{200, j} <138–20, G _{200, j} <168–25, B _{200, j} <148–23, GREY _{200, j} <151–20, and calculate the sum and count of each type of pixel. Experiments show that when i=200, redSum ₂₀₀ =6343, redCount ₂₀₀ =77, greenSum ₂₀₀ =7757, greenCount ₂₀₀ =72, blueSum ₂₀₀ =6763, blueCount ₂₀₀ =74, graySum ₂₀₀ =7286, grayCount ₂₀₀ =77.

Step 4, correct the binarization result of I[i][j] according to the calculation result of step 3:

1) If redCount _i is 0, set the i-th row of pixels as white (255); if redCount _i is not 0, make redAvgi equal to the average value of all red pixels, that is, redAvg _i = redSum _i /redCount _i ; traverse The entire row of pixels, if R _ij <redAvg _i +β, then set I[i][j] to black (0);

2) Continue to modify the binarization result according to the G _ij , B _ij , and GREY _ij channel information. Unlike the r channel, only the original white pixel value is corrected to black.

For example, when i=200, the pixel statistical data obtained in step 3 can be used to binarize the pixel points in the 200th row, that is, set the pixel value to black (0) or white (255).

Experiments show that the average value of red pixels redAvg ₂₀₀ =82, the average value of green pixels greenAvg ₂₀₀ =107, the average value of blue pixels blueAvg ₂₀₀ =91, and the average value of gray pixels greyAvg ₂₀₀ =94.

According to the red channel information, consider the pixel point of R _200,j <82+18 as the foreground, set its pixel to black, otherwise set it to white;

According to the green, blue, gray value and other information, the pixels in the 200th row are corrected three times. Set the white pixels that satisfy G _{200, j} <107+9, B _{200, j} <91+8, G _{200, j} <94+8 to black, and gradually correct the error of the first binarization.

Step 5, if i<499, i=i+1; return to step 2, and perform channel-based binarization processing on the next row of pixels. For example, when i=200, it is necessary to return to step 2 for the binarization processing of the next line; if i=499, it means that the binarization processing of the entire image is completed. The image of image 4 after binarization processing is shown in Figure 4.

The above description is a preferred embodiment of the present invention, and it should be pointed out that for those skilled in the art, without departing from the principle of the present invention, some improvements and modifications can also be made, and these improvements and modifications are also considered Be the protection scope of the present invention.

Claims (3)

1. an image binaryzation method based on passage, it is characterised in that comprise the following steps:

S1: for needing to carry out the image of binary conversion treatment, image being considered as a picture element matrix I, the width of image is designated as w, Highly it is designated as h；

S2: line by line to picture element matrix I process, scans the pixel of a line every time, collects statistical pixel point when processing the i-th row Three channel value red, blue, green of I [i] [j] and the information of gray value, be designated as R respectively_ij、G_ij、B_ijAnd GREY_ij, wherein GREY_ijIt is gray value, GREY_ij=(R_ij+G_ij+B_ij)/3, and calculate the meansigma methods of these four amounts in the i-th row pixel, it may be assumed that

redAvg_i=(∑_0≤j<wR_ij)/w,

greenAvg_i=(∑_0≤j<wG_ij)/w,

blueAvg_i=(∑_0≤j<wBij)/w,

greyAvg_i=(∑_0≤j<wGREY_ij)/w；

S3: the i-th row each passage pixel value is entered by each passage of the i-th row pixel according to meansigma methods and the threshold value of respective channel Row statistics also carries out binary conversion treatment:

S31: the redness of pixel, green, blue each channel value summation and gray value summation are designated as redSum respectively_i, greenSum_i, blueSum_i, greySum_i；Satisfactory redness, green, blueness, the pixel number of gray value are remembered respectively For redCount_i, greenCount_i, blueCount_i, greyCount_i；It is all initialized as 0；

S32: to red channel R_iProcess, use redAvg_i+ α, as cut off value, travels through the i-th row pixel, if R_ij< redAvg_i+ α, then it is assumed that I [i] [j] is red pixel point, and red color channel value is added to summation redSum_i, by redCount_i Add 1；Use identical method to G_i、B_iAnd GREY_iProcess, obtain greenSum_i、greenCount_i、blueSum_i、 blueCount_i、greySum_i, greyCount_i, wherein α is 1/10th of graphics standard variance；

S33: if redCount_iIt is 0, the i-th row pixel is set to white entirely；If redCount_iIt is not 0, makes redAvg_iIt is equal to The meansigma methods of all red pixel points, i.e. redAvg_i=redSum_i/redCount_i；Traversal entire row of pixels point, if R_ij< redAvg_i+ β, then arranging I [i] [j] is black；Wherein β is 1/10th of graphics standard variance；

S34: by red channel R_iReplace with G respectively_ij、B_ij、GREY_ijChannel information, repeats step S32-S33, revises binaryzation As a result, unlike red channel, only original white pixel value is modified to black；

S4: if i is < h-1, i=i+1；Then repeat step S2-S3, next line pixel is carried out binary conversion treatment based on passage.

2. image binaryzation method based on passage as claimed in claim 1, it is characterised in that step S2 be from top to bottom by Described picture element matrix is scanned by row.

3. image binaryzation method based on passage as claimed in claim 1, it is characterised in that step S32 is from left to right I-th row pixel is traveled through.