CN109034154A - The extraction and recognition methods of Invoice Seal duty paragraph - Google Patents

Abstract

The invention discloses a method for extracting and identifying a tax number of a special invoice stamp, which mainly solves the problems of poor efficiency and low precision in obtaining the tax number in the existing method. The implementation steps include: 1) obtaining the RGB image containing the invoice special stamp; 2) performing Laplace sharpening on the RGB image and converting it to the YCrCb color space; 3) performing adaptive threshold processing on the Cr channel of the YCrCb color space , only keep the content of the invoice special stamp; 4) Obtain the outer contour ellipse image of the invoice special stamp through Canny edge detection; 5) Find the center point by calculating the centroid coordinates of the ellipse image; 6) Intercept only the tax number according to the center point Rectangular area; 7) Divide the tax number in the rectangular area into single numbers; 8) Perform template matching on each number to identify the tax number. The invention effectively improves the acquisition speed of the tax number of the special invoice stamp, reduces the identification error, and can be used for fast and accurate extraction and identification of the tax number.

Description

Method for extracting and identifying the tax number of the special seal of the invoice

technical field

The invention belongs to the technical field of digital image processing, and further relates to character recognition technology, in particular to a method for extracting and identifying tax numbers of special invoice stamps, which can be used to obtain tax registration number information.

Background technique

The special seal for invoices refers to the special seals that are engraved by enterprises, units and individual industrial and commercial households that use invoices in accordance with the regulations of the tax authorities. They are stamped on the invoices when they receive purchases or issue invoices. The characters of the special seal for invoices and the tax registration number, where the tax registration number is referred to as the tax number. A tax ID is a tax “identity card” given by the issuing authority. A unified tax ID system is conducive to strengthening tax registration, standardizing tax collection and management, strengthening tax file management and tax source management, and promoting taxation according to law. Therefore, the extraction and identification of the tax number is particularly important.

At this stage, there are mainly two methods for extracting and identifying tax ID numbers: one is the traditional manual identification method, which manually identifies the tax ID number and then inputs it into the computer. This method has low efficiency, large errors, and excessive labor and time costs. The second is the extraction and recognition method based on deep learning. A classifier is trained through a large number of known training samples, and then the recurrent neural network is used to locate the preprocessed image digits, segment the image with digits, and finally segment the image. The image is sent to the trained classifier for classification. This method can effectively improve the problem of large error and high cost of manual recognition. However, both the cyclic neural network and the classifier used need to extract the specified features of the image as input, so The amount of calculation is large, and there are still shortcomings of slow extraction and recognition speed.

Contents of the invention

The purpose of the present invention is to address the above-mentioned deficiencies in the prior art, and propose a method for extracting and identifying the tax number of the special invoice stamp, by calculating the centroid of the ellipse image, quickly locate the center coordinates of the special stamp of the invoice, and then select the area where the tax number is located, Then segment a single digital image, perform template matching on it in turn, and quickly and accurately identify the tax number in the special invoice stamp, thereby improving the recognition efficiency and accuracy of the tax number in the special invoice stamp.

In order to achieve the above object, the technical scheme that the present invention takes is:

(1) Obtain the RGB image containing the special stamp for the invoice;

(2) Laplacian sharpening is carried out to the acquired RGB image, and then it is converted to the YCrCb color space;

(3) Separate the channels of the YCrCb color space, and perform adaptive threshold processing on the separated Cr channel, convert the image into a binary image, and then remove the noise of the binary image through dilation and corrosion operations, and obtain only the invoice-specific stamp The binary map;

(4) Carry out Canny edge detection on the binary image that only contains the special invoice stamp, and select the outermost contour, that is, the outer contour ellipse image of the special invoice stamp;

(5) Calculate the centroid coordinates of the outer contour ellipse image, i.e. the coordinates of the center point of the special invoice stamp;

(6) According to the coordinates of the center point, a rectangular area containing only the tax number is intercepted;

(7) Divide the tax code in the rectangular area into a single number;

(8) Make a search template T for numbers 0 to 9, adjust the size of a single digital image to make it consistent with the size of the number in the search template T, and send the adjusted single number as the searched image S to the prepared number in sequence In the search template, the correlation method is used to match the numbers, and the tax ID is identified according to the matching result.

Compared with the prior art, the present invention has the following advantages:

First, because the present invention uses the zero-order and first-order geometric moments to obtain the center point coordinates of the ellipse of the special invoice seal, and then accurately locates the position of the tax number according to the center point coordinates, there is no need to train the network and extract image features, thereby reducing the The amount of calculation for tax number positioning improves the positioning accuracy.

Second, because the present invention uses template matching to identify the digital part of the invoice special chapter, it does not need to train a classifier and extract image features, thereby speeding up the recognition of numbers, improving the extraction and recognition of tax Number of efficiency.

Description of drawings

Fig. 1 is the realization flowchart of the present invention;

Figure 2 is an RGB image containing a special stamp for invoices;

Figure 3 is a binary image containing only the content of the invoice special seal;

Figure 4 is the outline image of the invoice special seal;

Figure 5 is an image of the outer contour and center point of the invoice special seal;

Figure 6 is a rectangular binary map containing only tax codes;

Figure 7 is the image obtained after color inversion of the rectangular binary image containing only tax codes.

Detailed ways

In order to enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific embodiments.

Referring to Fig. 1, the method for extracting and identifying the tax number of the special invoice stamp includes the following steps:

Step 1. Obtain an RGB image containing the invoice-specific stamp, as shown in Figure 2.

Step 2. Laplacian sharpen the acquired RGB image and then convert it to YCrCb color space.

First, Laplacian sharpening is performed on the RGB image obtained in step 1, as follows: reduce the gray value of the center pixel that is lower than the average gray value of the neighborhood pixels, and lower the gray value of the center pixel that is higher than the average gray value The gray value of the central pixel increases. That is, when the gray level of the central pixel in the neighborhood is lower than the average gray level of other pixels in its area, reduce the gray value of the central pixel; when the gray level of the central pixel in the neighborhood is higher than the average gray level of other pixels in its neighborhood Increase the gray value of this center pixel when the brightness is higher. After the image is sharpened, the invoice-specific stamp part is easier to extract.

Then, the sharpened RGB image is converted to the YCrCb color space by the following formula:

Y＝(B×1868+G×9617+R×4899+8192)÷16384;

Cr＝((B-Y)×9241+8192)÷16384+128;

Cb=((R-Y)×11682+8192)÷16384+128;

Among them, "R" indicates the value of the R channel in the RGB image, "G" indicates the value of the G channel in the RGB image, "B" indicates the value of the B channel in the RGB image, and "Y" indicates the value of the Y channel in the YCrCb image, "Cr" indicates the value of the Cr channel in the YCrCb image, and "Cb" indicates the value of the Cb channel in the YCrCb image.

The YCrCb color space contains three channels, among which: "Y" represents brightness, that is, the grayscale value; "Cr" and "Cb" both represent chroma, "Cr" refers to the red chroma component, and "Cb" refers to the blue chroma portion.

Step 3. Perform adaptive threshold processing on the Cr channel of the YCrCb color space, and only keep the content of the invoice special chapter.

Separate the Cr channel from the YCrCb color space, perform adaptive threshold processing on the Cr channel, convert the image into a binary image, and then perform expansion and corrosion operations on it to remove noise, and obtain a binary image that only retains the special stamp for invoices, as shown in the figure 3.

Step 4. Obtain the outer contour ellipse image of the invoice special stamp through Canny edge detection.

Use the following steps to perform Canny edge detection on the binary image that only retains the special chapter of the invoice:

1) Use a Gaussian filter to smooth the image and filter out noise;

2) Calculate the gradient strength and direction of each pixel in the image;

3) Use non-maximum value suppression to eliminate the spurious response caused by edge detection;

4) Identify real and potential edges using dual-threshold detection;

5) Edge detection is finally completed by suppressing isolated weak edges.

After the edge detection is completed, the outermost contour is selected, that is, the outer contour ellipse image of the invoice special stamp, as shown in Figure 4.

Step 5. Find the center point of the invoice-specific stamp by calculating the centroid coordinates of the ellipse image.

According to the national standard, the tax number of the special invoice stamp is at its center, so it is necessary to find the center point of the special stamp of the invoice to locate the position of the tax number. There are many ways to calculate the centroid coordinates of the outer contour ellipse image, such as geometric moments, Hough transform, etc., the present invention uses the geometric moment method to calculate and obtain the centroid coordinates of the outer contour ellipse image, and the specific steps are as follows:

The p+q order geometric moment m _pq of a discrete image is defined as:

Among them, x and y respectively represent the horizontal and vertical coordinates of the outer contour ellipse image; p represents the order of x, q represents the order of y, and both p and q are integers greater than or equal to 0; M represents the abscissa of the external contour ellipse image The boundary of x, N represents the boundary of the y-coordinate of the outer contour ellipse image, and both M and N are integers greater than or equal to 1;

Take p=0, q=0, and get the zero-order geometric moment m ₀₀ according to formula <1>:

Take p=1, q=0, and get the first-order geometric moment m ₁₀ of x according to formula <1>:

Take p=0, q=1, and get the first-order geometric moment m ₀₁ of y according to formula <1>:

Using the zero-order and first-order geometric moments, the coordinates of the center of mass of the ellipse image of the outer contour of the invoice special seal are obtained by the following formula

The center point of the outer contour of the special seal for invoices is shown in Figure 5.

If the Hough transform is used to calculate the centroid coordinates of the outer contour ellipse image in this step, the following steps are required:

Represent a quadratic curve in the plane as:

Ax ² +2Bxy+Cy ² +2Dx+2Ey+1＝0

Among them, A, B, C, D, and E represent the five parameters to be obtained, and x and y represent the horizontal and vertical coordinates of the quadratic curve.

Calculate the above five parameters by using Hough transform, if B ² -AC<0, then the quadratic curve is an ellipse, and then solve the center coordinates of the ellipse according to the ellipse parameter equation.

Step 6. According to the coordinates of the center point, intercept the rectangular area containing only the tax number.

According to the sample standard of the special invoice stamp, the tax number is in the middle of the seal, and the smallest and complete rectangular area containing the tax number can be taken out according to the coordinates of the center point and the adaptive threshold, that is, the rectangular binary image that only contains the tax number, as shown in the figure 6.

Step 7. Split the tax ID number into single digits.

Invert the color of the rectangular area image extracted in step 6, and obtain the image after color inversion, as shown in Figure 7, and then segment the inverted image according to the adaptive threshold value, and segment each number. Splits tax ID numbers in a rectangular area into individual digits.

Step 8. Perform template matching on each number to identify the tax ID number.

Make a search template T for numbers 0 to 9, adjust the size of a single digital image to make it consistent with the size of the number in the search template T, and send the adjusted single number as the searched image S to the prepared digital search template in turn , use the correlation method to match the numbers, and identify the tax number according to the matching result.

The specific process of matching numbers by correlation method is as follows:

(8.1) Assuming that the search template T has m×n pixels, and the searched image S has W×H pixels, superimpose T on S and perform translation, the S area covered by T is the sub-image S _ij , According to the following formula, the similarity D(i,j) between the search template T and the subgraph S _ij is obtained:

Among them, i is the abscissa of the upper left corner of the subgraph on the searched graph S, j is the ordinate of the upper left corner of the subgraph on the searched graph S, and 1≤i≤Wm, 1≤j≤Hn, S _ij ( m, n) is the pixel value of S _ij at coordinates (m, n), and T(m, n) is the pixel value of T at coordinates (m, n);

(8.2) Normalize D(i,j) to get the correlation coefficient R(i,j) of template matching:

(8.3) The correlation coefficient R(i,j)=1 means that the template and the subgraph are completely consistent. Since there will be a certain deviation, we stipulate here that if R(i,j) is greater than or equal to 0.9, it means that the template and the subgraph are consistent. Determine whether the value of the correlation coefficient R(i,j) is greater than or equal to 0.9. If so, the subgraph is consistent with the template, complete the matching search, and determine the number corresponding to the picture; otherwise, the subgraph is inconsistent with the template, and continue to iteratively search until it is found A template consistent with the subgraph, to complete the matching search.

Parts not described in detail in the present invention belong to the common knowledge of those skilled in the art.

The above description is only an embodiment of the present invention, and does not constitute any limitation to the present invention. Obviously, for those skilled in the art, after understanding the content and principle of the present invention, it is possible without departing from the principle of the present invention. Various modifications and changes in form and details are made, but these modifications and changes based on the idea of the present invention are still within the protection scope of the claims of the present invention.

Claims (9)

1. A method for extracting and identifying tax ID numbers of special stamps for invoices, comprising the following steps: (1) Obtain the RGB image containing the special stamp for the invoice; (2) Laplacian sharpening is carried out to the acquired RGB image, and then it is converted to the YCrCb color space; (3) Separate the channels of the YCrCb color space, and perform adaptive threshold processing on the separated Cr channel, convert the image into a binary image, and then remove the noise of the binary image through dilation and corrosion operations, and obtain only the invoice-specific stamp The binary map; (4) Carry out Canny edge detection on the binary image that only contains the special invoice stamp, and select the outermost contour, that is, the outer contour ellipse image of the special invoice stamp; (5) Calculate the centroid coordinates of the outer contour ellipse image, i.e. the coordinates of the center point of the special invoice stamp; (6) According to the coordinates of the center point, a rectangular area containing only the tax number is intercepted; (7) Divide the tax code in the rectangular area into a single number; (8) Make a search template T for numbers 0 to 9, adjust the size of a single digital image to make it consistent with the size of the number in the search template T, and send the adjusted single number as the searched image S to the prepared number in sequence In the search template, the correlation method is used to match the numbers, and the tax ID is identified according to the matching result. 2. The method according to claim 1, characterized in that: in the step (2), carry out Laplace sharpening processing to the RGB image obtained, and the specific processing method is: the gray value lower than the neighborhood pixel average gray value The gray value of the central pixel decreases, and the gray value of the central pixel higher than the average gray value of the neighboring pixels is increased. 3. method according to claim 1, is characterized in that: in step (2), the RGB image after sharpening is converted to YCrCb color space by following formula: Y＝(B×1868+G×9617+R×4899+8192)÷16384; Cr＝((B-Y)×9241+8192)÷16384+128; Cb=((R-Y)×11682+8192)÷16384+128; Among them, "Y" represents the value of the Y channel in the YCrCb image, "Cr" represents the value of the Cr channel in the YCrCb image, "Cb" represents the value of the Cb channel in the YCrCb image, and "R" represents the value of the R channel in the RGB image, "G" indicates the value of the G channel in the RGB image, and "B" indicates the value of the B channel in the RGB image. 4. The method according to claim 1, characterized in that: in the step (4), the Canny edge detection is carried out to the binary image that only includes the special seal for invoices, specifically through the following steps: (4.1) Utilize Gaussian filter to denoise the binary image; (4.2) Calculate the gradient strength and direction of each pixel in the image after denoising; (4.3) Use non-maximum value suppression to eliminate the spurious response caused by edge detection; (4.4) Identify real and potential edges using dual-threshold detection; (4.5) Edge detection is finally accomplished by suppressing isolated weak edges. 5. The method according to claim 1, characterized in that: the centroid coordinates of the outer contour ellipse image in step (5) are calculated by geometric moments or Hough transform. 6. The method according to claim 5, characterized in that: calculate the centroid coordinates of the outer contour ellipse image by geometric moments, and the concrete steps are as follows: The p+q order geometric moment m _pq of a discrete image is defined as: Among them, x and y respectively represent the horizontal and vertical coordinates of the outer contour ellipse image; p represents the order of x, q represents the order of y, and both p and q are integers greater than or equal to 0; M represents the abscissa of the external contour ellipse image The boundary of x, N represents the boundary of the y-coordinate of the outer contour ellipse image, and both M and N are integers greater than or equal to 1; Take p=0, q=0, and get the zero-order geometric moment m ₀₀ according to formula <1>: Take p=1, q=0, and get the first-order geometric moment m ₁₀ of x according to formula <1>: Take p=0, q=1, and get the first-order geometric moment m ₀₁ of y according to formula <1>: Using the zero-order and first-order geometric moments, the coordinates of the center of mass of the ellipse image of the outer contour of the invoice special seal are obtained by the following formula 7. The method according to claim 1, characterized in that in step (6), the rectangular area containing only the tax code is the smallest and complete area containing the tax code extracted according to the coordinates of the center of mass and the adaptive threshold. 8. The method according to claim 1, characterized in that: in step (7), the tax code in the rectangular area is divided into single numbers, including the following operations: (7.1) Reverse the color of the image in the rectangular area to obtain the image after inversion; (7.2) Segment the reversed image according to the adaptive threshold to obtain a single digital image. 9. method according to claim 1 is characterized in that: adopt correlation method to match the specific process of number as follows in the step (8): (8.1) Assuming that the search template T has m×n pixels, and the searched image S has W×H pixels, superimpose T on S and perform translation, the S area covered by T is the sub-image S _ij , According to the following formula, the similarity D(i,j) between the search template T and the subgraph S _ij is obtained: Among them, i is the abscissa of the upper left corner of the subgraph on the searched graph S, j is the ordinate of the upper left corner of the subgraph on the searched graph S, and 1≤i≤Wm, 1≤j≤Hn, S _ij ( m, n) is the pixel value of S _ij at coordinates (m, n), and T(m, n) is the pixel value of T at coordinates (m, n); (8.2) Normalize D(i,j) to get the correlation coefficient R(i,j) of template matching: (8.3) Determine whether the value of the correlation coefficient R(i, j) is greater than or equal to 0.9, if so, the subgraph is consistent with the template, complete the matching search, and determine the number corresponding to the picture; otherwise, if the subgraph is inconsistent with the template, continue the iterative search, Until a template consistent with the subgraph is found, the matching search is completed.