CN109426814B - Method, system and equipment for positioning and identifying specific plate of invoice picture - Google Patents

Abstract

The invention relates to a positioning and identification method, system and equipment for a specific plate of an invoice picture. The method for locating and identifying the specific section of the invoice picture includes: obtaining the invoice picture; The text content of the information picture is recognized to obtain the invoice information to be recognized. Before the recognition, the target is directly positioned and cut, so the invoice information corresponding to each recognition content is clear and clear, and it is innovative to realize the automatic recognition of each invoice, which fills the gap in this technical field and has broad application prospects. . It fundamentally solves the disadvantage that the existing image recognition algorithm can only recognize each character in the invoice picture indiscriminately, but cannot distinguish the required invoice information. The acquisition of invoice information is truly automated.

Description

A positioning and identification method, system, and equipment for a specific section of an invoice image

technical field

The invention relates to invoices, in particular to a positioning and identification method, system and equipment for a specific plate of an invoice picture.

Background technique

Processing invoice reimbursement is a very important task for the company's financial department. Staff need to obtain information such as the invoice number sequence, company name, invoice date, amount, and tax amount in each invoice for further verification and other work. Heavy work.

Although the traditional image recognition algorithm can identify every word and every symbol in the invoice picture line by line without distinction, it cannot recognize which field is the invoice number, which field is the date, and which field is the amount. wait.

Therefore, the acquisition of invoice information can only be done manually.

Contents of the invention

In order to solve the above-mentioned technical problems, the object of the present invention is to provide a method, system and equipment for locating and identifying a specific section of an invoice image.

According to one aspect of the present invention, a method for locating and identifying a specific section of an invoice image is provided, including:

Get the picture of the invoice;

Determine the invoice information to be identified;

According to the invoice information to be identified, locate and segment the area of the invoice picture where it is located, and obtain the image of the invoice information to be identified;

Identify the text content of the image of the invoice information to be identified, and obtain the invoice information to be identified.

The present invention obtains the pictures of the left and right invoices to be processed by any simple method, and then intercepts and straightens the effective areas of the invoices in all pictures at various positions, various rotation angles, and various brightness conditions through some image processing methods, and then combines Logical processing, using Canny contour detection, image text detection, color detection, and extraction of "rows and columns" to accurately locate each area to be recognized in a picture, and intercept them into fragments for recognition separately. The original manual method is replaced by automatically obtaining invoice information. Not only the accuracy is high, but also the efficiency is greatly improved.

Further, the invoice information to be identified includes invoice number sequence, company name, invoice date, amount and tax amount.

Further, after obtaining the invoice image, it includes intercepting the effective area of the invoice.

Further, before intercepting the effective area of the invoice, it includes: aligning the invoice picture.

Further, the straightened invoice pictures include:

Use openCV's HoughLine algorithm to detect all the lines in the invoice image;

Calculate the deflection angle for each line;

The rotation angle that appears most times is the deflection angle of the picture;

Rotate the picture according to the deflection angle to straighten the picture of the invoice.

Further, intercepting the effective area of the invoice includes:

Extract the outline of the effective area of the invoice in the invoice image;

A picture of the effective area of the invoice is obtained based on the contour segmentation.

Further, after obtaining the picture of the effective area of the invoice, the execution step includes using the SIFT algorithm to determine whether the position of the invoice head is above the picture, otherwise, flipping the picture upside down. Among them, SIFT is a computer vision algorithm used to detect and describe local features in images.

Further, the outline of the effective area of the invoice in the invoice image is extracted, including:

Calculate the sum of the pixel values of each pixel column and the sum of the pixel values of each pixel row of the invoice image respectively to obtain two vectors;

Multiply two vectors to get a two-dimensional array;

Convert this two-dimensional array into the corresponding grayscale image;

Filter and remove the horizontal and vertical lines in the invoice image to obtain the outline of the effective area of the invoice.

Specifically:

Calculate the sum of the pixel values of each pixel column and the sum of the pixel values of each pixel row of the entire invoice image with no background removed to obtain two vectors (the sum of the pixel rows or columns calculated here is for It is calculated by including a whole picture of the background, because the program does not know where the outline of the effective area of the invoice is, and the program needs to find the outline of the effective area of the invoice by calculating the pixel row and column sum, multiplying, and removing horizontal lines. where);

Multiply two vectors to get a two-dimensional array;

Convert this two-dimensional array into a corresponding grayscale image (normalize all values to the range of 0 to 255 in proportion, each value from 0 to 255 represents from white to different levels of gray to black in turn grayscale value);

Through the cv2.morphologyEx function in the openCV toolkit (directed filtering of a certain form in the picture), the horizontal and vertical lines in the picture are directional filtered (by changing the parameters to determine which form to filter), and the horizontal and vertical lines in the picture are removed. line to get the outline of the effective area of the invoice.

The above method ensures that the effective area of the invoice can be extracted without being affected by interference factors such as the position angle of the effective area of the invoice in the photo, the degree of light and shade, light and shadow of the picture, and the effective area of the invoice can be accurately intercepted.

Further, the positioning and segmentation of the area where the invoice information is to be identified includes:

Use the Canny algorithm to obtain the area matching the preset positioning template and the vertex coordinates of the matching area;

According to the vertex coordinates of the matching area, obtain the image of the area where the invoice information to be identified is located;

Extract the pixel of the invoice information text to be recognized in the image of the area, and perform reverse binarization processing;

Calculate the sum of the pixel values of the rows and columns of the image in the area, compare it with the preset threshold value for distinguishing characters and gaps, and judge and segment the image of the invoice information to be recognized.

Further, the Canny algorithm is used to obtain the area matching the preset positioning template and the vertex coordinates of the matching area, including:

Capture some pictures of the area where the invoice information to be identified is located;

Scale the part of the picture to get multiple zoomed pictures;

Input multiple zoom images and preset positioning templates into the Canny algorithm for calculation, and obtain the area matching the preset positioning template and the vertex coordinates of the matching area.

The contour matching of multiple zoom images and the positioning template is performed to locate the matching area, and the positioning accuracy is high.

After locating the matching area and obtaining the vertex coordinates of the matching area, according to the positional relationship between the matching area and the image of the area where the invoice information to be identified is obtained, locate and acquire the image of the area where the invoice information to be identified is obtained.

Wherein, the scaling parameter is 0.8-1.2.

Further, calculating the sum of the pixel values of the row and column of the image in the area, comparing it with the threshold value, judging and segmenting the invoice information image to be identified includes:

Calculate the sum of the pixel values of each pixel row of the region picture, judge and segment the row region picture where the invoice information to be identified is located according to the threshold value of the pixel row;

Calculate the sum of the pixel values of each pixel column in the row area picture, judge and segment the invoice information picture to be recognized according to the threshold value of the pixel column.

Among them, the above threshold needs to be determined according to the amount of text in each row or column of the picture and the amount of noise in the picture. In actual operation, the sum of the pixel values of each pixel row or column is first output, and then the developer checks if there is no text. The approximate critical value of the value of the sum of pixels in the pixel row or column and the pixel row or column with text is the threshold value.

Wherein, the threshold value of the pixel row is 14000, and the threshold value of the pixel column is 4000.

According to another aspect of the present invention, a positioning and identification system for a specific block of an invoice image is provided, including:

Acquisition unit for obtaining invoice pictures;

According to the invoice information to be identified, locate and segment the area of the invoice image where it is located, and obtain the positioning unit of the image of the invoice information to be identified;

Recognize the text content of the picture of the invoice information to be recognized, and obtain the recognition unit of the invoice information to be recognized.

Further, the invoice information to be identified includes invoice number sequence, company name, invoice date, amount and tax amount.

The system is based on the positioning and identification method of the above-mentioned specific section of the invoice picture, which intercepts the effective area of the invoice, locates and divides the area where the invoice information to be identified is located, and identifies the text content of the image of the invoice information to be identified, and obtains the invoice information to be identified, etc. The steps are as described in the positioning and identification methods of the specific section of the invoice image.

According to another aspect of the present invention, a device for locating and identifying a specific block of an invoice image is provided, comprising a computer-readable medium storing a computer program, the program being run to perform:

Get the picture of the invoice;

Determine the invoice information to be identified;

According to the invoice information to be identified, locate and segment the area of the invoice picture where it is located, and obtain the image of the invoice information to be identified;

Identify the text content of the image of the invoice information to be identified, and obtain the invoice information to be identified.

The device is based on the positioning and identification method of the above-mentioned specific section of the invoice picture, which intercepts the effective area of the invoice, locates and divides the area where the invoice information to be identified is located, and identifies the text content of the image of the invoice information to be identified, and obtains the invoice information to be identified, etc. The steps are as described in the positioning and identification methods of the specific section of the invoice image.

Compared with the prior art, the present invention has the following beneficial effects:

1. The positioning and identification method of the specific plate of the invoice picture of the example of the present invention, according to the invoice information to be identified, locate and segment the area of the invoice picture where it is located, and obtain the image of the invoice information to be identified; identify the text content of the image of the invoice information to be identified , to obtain the invoice information to be recognized. Before the identification, the target is directly positioned and cut, so the invoice information corresponding to each identification content is clear and clear. It is innovative and realizes the automatic identification of each invoice, which fills the gap in this technical field and has broad application prospects. .

2. The method for locating and identifying the specific plate of the invoice picture in the example of the present invention can extract the effective area of the invoice without being affected by interference factors such as the position angle of the effective area of the invoice in the photo, the degree of light and shade, the light and shadow of the picture, and the invoice can be accurately intercepted The effective area; and further detect the deflection angle of the invoice, and straighten the picture. Furthermore, combined with logical processing, the methods of Canny contour detection, image text detection, color detection, and row and column extraction are used to accurately locate each area to be recognized in a picture, and intercept each area to be recognized separately. Ensure accurate positioning and accurate identification.

3. The positioning and identification system of the specific plate of the invoice picture in the example of the present invention obtains the invoice picture through the acquisition unit; through the positioning unit, according to the invoice information to be identified, the area of the invoice picture where it is located is positioned and segmented to obtain the invoice information picture to be identified ; Identify the text content of the invoice information picture to be identified by the identification unit, and obtain the invoice information to be identified. Since the information of the invoice to be identified is clear and the position of the invoice is clear, the invoice information to be identified is first determined through implementation, and then the invoice information is determined. The area where the information is located is located and cut to obtain a picture with information content to be identified, and then the corresponding invoice information is obtained by identifying the image, so the identification information corresponding to each identification process is clear, which fundamentally solves the problem of existing image identification algorithms. It can only identify each character in the invoice picture without distinction, but cannot distinguish the required invoice information. The acquisition of invoice information is truly automated.

4. The positioning and identification equipment of the specific plate of the invoice image in the example of the present invention is stored and operated to perform the following procedures: obtain the invoice image; determine the invoice information to be identified; according to the invoice information to be identified, identify the area of the invoice image Locate and segment to obtain the image of the invoice information to be identified; identify the text content of the image of the invoice information to be identified to obtain the invoice information to be identified. Only the invoice picture is needed, and the invoice information can be automatically identified by running the above program without manual identification, which not only ensures the accuracy, but also greatly improves the efficiency.

Description of drawings

Fig. 1 is a schematic diagram of an initial invoice picture obtained in Embodiment 1 of the present invention;

Fig. 2 is a schematic diagram of an invoice picture in Embodiment 1 of the present invention;

Fig. 3 is a schematic diagram of an effective area for straightening an invoice picture in Embodiment 1 of the present invention;

Fig. 4 is a schematic diagram of an invoice valid area that is straightened and not reversed according to Embodiment 1 of the present invention;

Fig. 5 is a schematic diagram of the name of the company whose invoice information is to be identified according to an embodiment of the present invention;

6 is a schematic diagram of a positioning template according to an embodiment of the present invention;

Fig. 7 is a zoomed-in diagram with the highest degree of matching with the positioning template according to Embodiment 1 of the present invention;

Fig. 8 is a schematic diagram of the area where the company name of the invoice information to be identified is located according to Embodiment 1 of the present invention;

Fig. 9 is a schematic diagram of the line where the company name of the invoice information to be identified is located according to the embodiment of the present invention;

Fig. 10 is a schematic diagram of a picture of the company name of the invoice information to be recognized in Embodiment 1 of the present invention.

Detailed ways

In order to better understand the technical solutions of the present invention, the present invention will be further described below in conjunction with specific examples.

Embodiment one:

This embodiment provides a method for locating and identifying a specific section of an invoice picture, taking a value-added tax invoice as an example, and the method includes:

S1. Extract and straighten the effective area of the invoice from the VAT invoice picture:

Invoice pictures are often taken by people under various uncertain actual conditions, so the position of the effective area of the invoice in the picture, the background pattern, light, clarity and the rotation angle of the invoice are not fixed. Such a situation is It is impossible to directly call the image recognition program to identify, so it is necessary to first locate the precise position of each target and intercept them into individual fragments for identification.

S1.1. Obtain the picture of the invoice, as shown in Figure 1;

S1.2. Read the invoice picture, use openCV's HoughLine algorithm to detect all the lines in the invoice picture; calculate the deflection angle of each line; find the most frequent rotation angle, which is the deflection angle of the picture; reverse according to the deflection angle Rotate the picture to obtain a straightened picture, as shown in Figure 2;

S1.3. For the straightened invoice image, first extract the contours of all details, calculate the sum of the pixel values of each pixel column, and then calculate the sum of the pixel values of each pixel row, and multiply the two vectors to form a two-dimensional array (It can be treated as a two-dimensional image); remove the horizontal and vertical lines in the image to completely divide the invoice area and background, obtain the outline of the effective area of the invoice image, intercept, and obtain the image of the effective area of the invoice, as shown in Figure 3 As shown, specifically:

Extract the outline of the valid area of the invoice that squares the invoice image:

Calculate the sum of the pixel values of each pixel column and the sum of the pixel values of each pixel row of the entire invoice image with no background removed to obtain two vectors (the sum of the pixel rows or columns calculated here is for It is calculated by including a whole picture of the background, because the program does not know where the outline of the effective area of the invoice is, and the program needs to find the outline of the effective area of the invoice by calculating the pixel row and column sum, multiplying, and removing horizontal lines. where);

Multiply two vectors to get a two-dimensional array;

Convert this two-dimensional array into a corresponding grayscale image (normalize all values to the range of 0 to 255 in proportion, each value from 0 to 255 represents from white to different levels of gray to black in turn grayscale value);

Through the cv2.morphologyEx function in the openCV toolkit (directed filtering of a certain form in the picture), the horizontal and vertical lines in the picture are directional filtered (by changing the parameters to determine which form to filter), and the horizontal and vertical lines in the picture are removed. Line to get the outline of the effective area of the invoice;

S1.4. Through the SIFT algorithm, find out whether the position of the invoice head is above the picture. If not, flip the picture up and down, and finally get an invoice picture with no deflection angle and no upside down, only the valid area of the invoice, as shown in Figure 4 .

S2. According to the invoice information to be identified, locate and segment the area of the invoice picture where it is located, and obtain the image of the invoice information to be identified:

S2.1. Determine the invoice information to be identified: company name, as shown in Figure 5;

S2.2. Determine the positioning template according to the invoice information to be identified, as shown in Figure 6, since the position of the invoice information to be identified in the invoice is fixed, different positioning templates can be determined according to different identification information;

S2.3 Use the positioning template to perform template matching on the entire picture, determine its general position in the invoice picture, and intercept part of the picture in the area where the invoice information to be identified is located. The present invention purposefully locates the concerned plate in the value-added tax invoice, and automatically detects it. filter out unwanted areas;

S2.4. Perform 10-step zooming on some pictures in the area where the company name information is located from 0.8 to 1.2 times the scale parameter, and obtain 10 zoomed pictures;

S2.5. Obtain these 10 pictures and the Canny contour map of the positioning template through calculation, and then find the area similar to the template through template matching from these 10 pictures, and calculate the matching degree at the same time, and keep the zoomed picture with the highest matching degree, as shown in the figure The four vertex coordinates of the position shown in 7 and the matching area;

S2.6. According to the four coordinates obtained above, frame the area of the four lines of content, as shown in Figure 8;

S2.7. Extract all the pixels of the company name text from the image in the framed area, and then perform reverse binarization (pixels with a gray value greater than zero are converted to white 255), and an image with a black background and gray characters is obtained . Calculate the sum of the pixel values of each pixel line, and distinguish between text lines and gaps based on a certain threshold value of 14,000. If the sum of pixel values is greater than 14,000, it is a text line, and the pixel value is less than 14,000. And segment the three lines with words in the image, intercept the first line for further target, as shown in Figure 9.

S2.8. Calculate the sum of the pixel values of each pixel column in the above figure, judge and segment the text area according to the threshold value of 4000, and then obtain the picture of the company name area, as shown in Figure 10;

S3. The text content of the invoice information picture to be recognized can use the current open source tools for recognizing printed Chinese character strings and number strings, such as Google's tesseract toolkit. By calling the recognition function of this toolkit in the program, the picture content can be directly output. Identifying means obtaining the invoice information to be identified.

Obtain the invoice information to be recognized.

The present invention can be used to accurately locate a specific target area of a value-added tax invoice, intercept target plates one by one into target fragments for identification, and can effectively assist in improving the accuracy of invoice identification.

A positioning and identification system for a specific block of an invoice picture in this embodiment includes:

Acquisition unit for obtaining invoice pictures;

According to the invoice information to be identified, locate and segment the area of the invoice image where it is located, and obtain the positioning unit of the image of the invoice information to be identified;

Recognize the text content of the picture of the invoice information to be recognized, and obtain the recognition unit of the invoice information to be recognized.

A device for locating and identifying a specific section of an invoice picture in this embodiment includes a computer-readable medium storing a computer program, and the program is run to execute:

Get the picture of the invoice;

Determine the invoice information to be identified;

According to the invoice information to be identified, locate and segment the area of the invoice picture where it is located, and obtain the image of the invoice information to be identified;

Identify the text content of the image of the invoice information to be identified, and obtain the invoice information to be identified.

Embodiment two:

The features of this embodiment that are the same as those of Embodiment 1 will not be described in detail. The features of this embodiment that are different from Embodiment 1 are:

In the method for locating and identifying a specific section of an invoice picture in this embodiment,

S2. According to the invoice information to be identified, locate and segment the area of the invoice picture where it is located, and obtain the image of the invoice information to be identified:

S2.1. Determine the invoice information to be identified: the invoice number sequence;

S2.2. Determine the positioning template according to the invoice information to be identified;

S2.3. Use the positioning template to perform template matching on the entire image, determine its general position in the invoice image, and intercept part of the image in the area where the invoice information is to be identified;

S2.4. Perform 10-step zooming on the part of the picture in the area where the invoice information to be identified is located from 0.8 to 1.2 times the scale parameter, and obtain 10 zooming pictures;

S2.5. Obtain these 10 pictures and the Canny contour map of the positioning template through calculation, and then find the area similar to the template through template matching from these 10 pictures, and calculate the matching degree at the same time, and keep the zoomed picture and matching area with the highest matching degree The four vertex coordinates of the position;

S2.6. According to the four coordinates obtained above, frame the area of the four lines of content;

S2.7. Extract all the pixels of the invoice information text to be recognized from the image in the framed area, and then perform reverse binarization (pixels with a gray value greater than zero are converted to white 255), and the background is black, and the text is gray. Image. Calculate the sum of the pixel values of each pixel row, and distinguish between text rows and gaps according to a certain threshold value. If the sum of pixel values is greater than the threshold value, the region is a text row, and the pixels smaller than the threshold value act as gaps. This judges and divides the line with words in the image, and intercepts the picture of the line where the target is located.

S2.8. Calculate the sum of the pixel values of each pixel column in the above figure, judge and segment the text area according to the threshold value, and then obtain the picture of the ticket number sequence area;

S3. Identify the text content of the image of the invoice information to be identified, and obtain the invoice information to be identified.

Embodiment three:

The features of this embodiment that are the same as those of Embodiment 1 will not be described in detail. The features of this embodiment that are different from Embodiment 1 are:

In the method for locating and identifying a specific section of an invoice picture in this embodiment,

S2. According to the invoice information to be identified, locate and segment the area of the invoice picture where it is located, and obtain the image of the invoice information to be identified:

S2.1. Determine the invoice information to be identified: date of invoice;

S2.2. Determine the positioning template according to the invoice information to be identified;

S2.3. Use the positioning template to perform template matching on the entire image, determine its general position in the invoice image, and intercept part of the image in the area where the invoice information is to be identified;

S2.4. Perform 15 steps of zooming in on the part of the picture in the area where the invoice information to be identified is located from 0.8 to 1.2 times the scale parameter, and obtain 15 zoomed pictures;

S2.5. Obtain these 15 pictures and the Canny contour map of the positioning template, and then find the area similar to the template through template matching from these 10 pictures, and calculate the matching degree at the same time, and keep the zoomed picture and matching area with the highest matching degree The four vertex coordinates of the position;

S2.6. According to the four coordinates obtained above, frame the area of the four lines of content;

S2.7. Extract all the pixels of the text of the invoice information to be recognized from the image in the framed area, and then perform reverse binarization to obtain an image with a black background and gray text. Calculate the sum of the pixel values of each pixel row, and distinguish between text rows and gaps according to a certain threshold value. If the sum of pixel values is greater than the threshold value, the region is a text row, and the pixels smaller than the threshold value act as gaps. This judges and divides the line with words in the image, and intercepts the picture of the line where the target is located.

S2.8. Calculate the sum of the pixel values of each pixel column in the above figure, judge and segment the text area according to the threshold value, and then obtain the picture of the billing date area;

S3. Identify the text content of the image of the invoice information to be identified, and obtain the invoice information to be identified.

Embodiment four:

The features of this embodiment that are the same as those of Embodiment 1 will not be described in detail. The features of this embodiment that are different from Embodiment 1 are:

In the method for locating and identifying a specific section of an invoice picture in this embodiment,

S2. According to the invoice information to be identified, locate and segment the area of the invoice picture where it is located, and obtain the image of the invoice information to be identified:

S2.1. Determine the invoice information to be identified: amount;

S2.2. Determine the positioning template according to the invoice information to be identified;

S2.3. Use the positioning template to perform template matching on the entire image, determine its general position in the invoice image, and intercept part of the image in the area where the invoice information is to be identified;

S2.4. Perform 10-step zooming on the part of the picture in the area where the invoice information to be identified is located from 0.8 to 1.2 times the scale parameter, and obtain 10 zooming pictures;

S2.5. Obtain these 10 pictures and the Canny contour map of the positioning template through calculation, and then find the area similar to the template through template matching from these 10 pictures, and calculate the matching degree at the same time, and keep the zoomed picture and matching area with the highest matching degree The four vertex coordinates of the position;

S2.6. According to the four coordinates obtained above, frame the area of the four lines of content;

S2.7. Extract all the pixels of the text of the invoice information to be recognized from the image in the framed area, and then perform reverse binarization to obtain an image with a black background and gray text. Calculate the sum of the pixel values of each pixel row, and distinguish between text rows and gaps according to a certain threshold value. If the sum of pixel values is greater than the threshold value, the region is a text row, and the pixels smaller than the threshold value act as gaps. This judges and divides the line with words in the image, and intercepts the picture of the line where the target is located.

S2.8. Calculate the sum of the pixel values of each pixel column in the above figure, judge and segment the text area according to the threshold value, and then obtain the picture of the amount area;

S3. Identify the text content of the image of the invoice information to be identified, and obtain the invoice information to be identified.

Embodiment five:

The features of this embodiment that are the same as those of Embodiment 1 will not be described in detail. The features of this embodiment that are different from Embodiment 1 are:

In the method for locating and identifying a specific section of an invoice picture in this embodiment,

S2. According to the invoice information to be identified, locate and segment the area of the invoice picture where it is located, and obtain the image of the invoice information to be identified:

S2.1. Determine the invoice information to be identified: tax amount;

S2.2. Determine the positioning template according to the invoice information to be identified;

S2.3. Use the positioning template to perform template matching on the entire image, determine its general position in the invoice image, and intercept part of the image in the area where the invoice information is to be identified;

S2.4. Perform 10-step zooming on the part of the picture in the area where the invoice information to be identified is located from 0.8 to 1.2 times the scale parameter, and obtain 10 zooming pictures;

S2.5. Obtain these 10 pictures and the Canny contour map of the positioning template through calculation, and then find the area similar to the template through template matching from these 10 pictures, and calculate the matching degree at the same time, and keep the zoomed picture and matching area with the highest matching degree The four vertex coordinates of the position;

S2.6. According to the four coordinates obtained above, frame the area of the four lines of content;

S2.7. Extract all the pixels of the text of the invoice information to be recognized from the image in the framed area, and then perform reverse binarization to obtain an image with a black background and gray text. Calculate the sum of the pixel values of each pixel row, and distinguish between text rows and gaps according to a certain threshold value. If the sum of pixel values is greater than the threshold value, the region is a text row, and the pixels smaller than the threshold value act as gaps. This judges and divides the line with words in the image, and intercepts the picture of the line where the target is located.

S2.8. Calculate the sum of the pixel values of each pixel column in the above figure, judge and segment the text area according to the threshold value, and then obtain the picture of the tax amount area;

S3. Identify the text content of the image of the invoice information to be identified, and obtain the invoice information to be identified.

The above description is only a preferred embodiment of the present application and an illustration of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solution formed by the specific combination of the above-mentioned technical features, but should also cover the technical solution formed by the above-mentioned technical features without departing from the inventive concept. Other technical solutions formed by any combination of or equivalent features thereof. For example, the features described above have similar functions to those disclosed (but not limited to) in this application.

Claims (11)

1. A method for positioning and identifying a specific section of an invoice picture, characterized in that it comprises: Get the picture of the invoice; Determine the invoice information to be identified; According to the invoice information to be identified, a positioning template is determined, and different information to be identified corresponds to different positioning templates, and the invoice information to be identified includes a sequence of ticket numbers, a company name, an invoice date, an amount and a tax amount; Use the positioning template to perform template matching on the invoice picture, determine its position in the invoice picture, and intercept part of the picture in the area where the invoice information is to be identified; Scaling part of the pictures in the area where the information to be identified is located to obtain a zoomed picture; Using the Canny algorithm to obtain the region where the zoom map matches the determined positioning template and the vertex coordinates of the matching region; According to the vertex coordinates of the matching area, obtain the image of the area where the invoice information to be identified is located; Extract the pixel of the invoice information text to be recognized in the image of the area, and perform reverse binarization processing; Calculate the sum of the pixel values of the row and column of the image in the area, compare it with the preset threshold value for distinguishing text and gaps, judge and segment the image of the invoice information to be identified; Identify the text content of the image of the invoice information to be identified, and obtain the invoice information to be identified. 2. The method for locating and identifying a specific section of an invoice image according to claim 1, wherein the step of executing after obtaining the invoice image includes intercepting the effective area of the invoice. 3. The method for locating and identifying a specific section of an invoice picture according to claim 2, wherein the method of intercepting the effective area of the invoice includes: aligning the invoice picture. 4. The positioning and identification method of the specific plate of the invoice picture according to claim 3, characterized in that, the straightening of the invoice picture comprises: Use openCV's HoughLine algorithm to detect all the lines in the invoice image; Calculate the deflection angle for each line; The rotation angle that appears most times is the deflection angle of the picture; Rotate the picture according to the deflection angle to straighten the picture of the invoice. 5. The method for locating and identifying a specific section of an invoice picture according to claim 3, wherein the intercepting the effective area of an invoice comprises: Extract the outline of the effective area of the invoice in the invoice image; A picture of the effective area of the invoice is obtained based on the contour segmentation. 6. The method for positioning and identifying the specific plate of the invoice picture according to claim 5, wherein extracting the outline of the effective area of the invoice in the invoice picture comprises: Calculate the sum of the pixel values of each pixel column and the sum of the pixel values of each pixel row of the invoice image respectively to obtain two vectors; Multiply two vectors to get a two-dimensional array; Convert this two-dimensional array into the corresponding grayscale image; Filter and remove the horizontal and vertical lines in the invoice image to obtain the outline of the effective area of the invoice. 7. The positioning and identification method of the specific plate of the invoice picture according to claim 6, wherein the step of executing after obtaining the picture of the effective area of the invoice includes passing through the SIFT algorithm to determine whether the position of the invoice head is above the picture, otherwise , flips the image upside down. 8. The positioning and identification method of the specific plate of the invoice picture according to claim 1, wherein the method for locating and identifying the specific plate of the invoice image is characterized in that, using the Canny algorithm to obtain the area matched with the preset positioning template and the vertex coordinates of the matching area, including: Capture some pictures of the area where the invoice information to be identified is located; Scale the part of the picture to get multiple zoomed pictures; Input multiple zoom images and preset positioning templates into the Canny algorithm for calculation, and obtain the area matching the preset positioning template and the vertex coordinates of the matching area. 9. The method for locating and identifying a specific plate of an invoice picture according to claim 1, wherein the sum of the pixel values of the row and column of the area picture is calculated, compared with a threshold value, judged and segmented out The picture of the invoice information to be recognized includes: Calculate the sum of the pixel values of each pixel row of the region picture, judge and segment the row region picture where the invoice information to be identified is located according to the threshold value of the pixel row; Calculate the sum of the pixel values of each pixel column in the row area picture, judge and segment the invoice information picture to be recognized according to the threshold value of the pixel column. 10. A positioning and identification system for a specific section of an invoice image, characterized in that it includes: Acquisition unit for obtaining invoice pictures; According to the invoice information to be identified, locate and segment the area of the invoice image where it is located, and obtain the positioning unit of the image of the invoice information to be identified; Recognize the text content of the image of the invoice information to be identified, and obtain the identification unit of the invoice information to be identified; The positioning and segmentation of the area where the invoice information to be identified includes the positioning unit is specifically used for: According to the invoice information to be identified, a positioning template is determined, and different information to be identified corresponds to different positioning templates, and the invoice information to be identified includes a sequence of ticket numbers, a company name, an invoice date, an amount and a tax amount; Use the positioning template to perform template matching on the invoice picture, determine its position in the invoice picture, and intercept part of the picture in the area where the invoice information is to be identified; Scaling part of the pictures in the area where the information to be identified is located to obtain a zoomed picture; Using the Canny algorithm to obtain the region where the zoom map matches the determined positioning template and the vertex coordinates of the matching region; According to the vertex coordinates of the matching area, obtain the image of the area where the invoice information to be identified is located; Extract the pixel of the invoice information text to be recognized in the image of the area, and perform reverse binarization processing; Calculate the sum of the pixel values of the rows and columns of the image in the area, compare it with the preset threshold value for distinguishing characters and gaps, and judge and segment the image of the invoice information to be recognized. 11. A device for locating and identifying a specific block of an invoice image, characterized in that it includes a computer-readable medium storing a computer program, and the program is run to execute: Get the picture of the invoice; Determine the invoice information to be identified; According to the invoice information to be identified, locate and segment the area of the invoice picture where it is located, and obtain the image of the invoice information to be identified; Identify the text content of the image of the invoice information to be identified, and obtain the invoice information to be identified; The positioning and segmentation of the area where the invoice information is to be identified includes: According to the invoice information to be identified, a positioning template is determined, and different positioning templates are determined for different information to be identified, and the invoice information to be identified includes a sequence of ticket numbers, a company name, an invoice date, an amount and a tax amount; Use the positioning template to perform template matching on the invoice picture, determine its position in the invoice picture, and intercept part of the picture in the area where the invoice information is to be identified; Scaling part of the pictures in the area where the information to be identified is located to obtain a zoomed picture; Using the Canny algorithm to obtain the region where the zoom map matches the determined positioning template and the vertex coordinates of the matching region; According to the vertex coordinates of the matching area, obtain the image of the area where the invoice information to be identified is located; Extract the pixel of the invoice information text to be recognized in the image of the area, and perform reverse binarization processing; Calculate the sum of the pixel values of the rows and columns of the image in the area, compare it with the preset threshold value for distinguishing characters and gaps, and judge and segment the image of the invoice information to be recognized.

Images