CN114821611A - A method for protecting private data in archive table images - Google Patents

Abstract

The invention discloses a method for protecting private data in a form image of an archive, which utilizes RGB color difference of a color image to fade contents such as a color seal and the like of the form image; scaling the image to a specific size, and obtaining an edge intensity image of the image through convolution operation; obtaining candidate horizontal line equations and candidate vertical line equations by utilizing Hough transformation; backtracking and searching other possible tracking tracks; the coordinate information of the utilization line is obtained from the file user, and the private part and the part needing to be utilized are automatically blurred in the file image. The method has high calculation speed, and can complete identification and privacy fuzzy processing within 1 second on a common office computer; and the anti-interference capability is strong, and the old file, the table with the color seal, the original correction trace and the table with extremely light lines can be accurately identified. In addition, the compatibility is good, and the tables with different length-width ratios, different row and column numbers and different header styles can be effectively identified and processed.

Description

A method for protecting private data in archive table images

technical field

The invention relates to the field of file utilization and image processing, in particular to an image recognition and processing algorithm of file forms.

Background technique

In order to respect the original record of the files and the authenticity of the content, the file management department will scan and save the original files, and when issuing file certificates and other utilization materials, the information of others in the images that have nothing to do with the file users will be masked or blurred. .

The current masking method is to manually open the original file image using image processing software such as Paint or PhotoShop, manually select the part that needs to be processed for privacy, process the image, save, print, sign and seal and deliver it to the file user. Among them, the operation of manual selection of privacy processing is relatively cumbersome.

Computer vision technology has long been used to recognize tables in document images. Traditional image recognition and processing technologies include image skew correction, image binarization, horizontal and vertical projection of table horizontal and vertical lines, and Hough change voting to obtain line segment equations. Then perform line segment tracking detection and other technologies. When the image background is clean and the table is simple and standardized, this kind of method has better effect. However, a large number of file tables have various interference factors such as complicated headers, outdated and gray, curved lines, handwriting and signatures covering the table images, etc., and it is difficult to obtain the table wireframe information by the above method.

Another branch of computer vision technology is machine learning technology based on deep neural networks. This technology uses multi-layer convolution layers, pooling layers, activation functions, loss functions and other methods to construct a set of self-learning large numbers of samples, from images. A learning mechanism from features such as low-level edges to features such as high-level structures. Disadvantages include: a large number of learning samples need to be prepared and marked, and high computing power is required for learning and reasoning calculations. Among them, in the early stage of machine learning, the workload of collecting and labeling a large number of files and table samples is very large. Even if a large amount of manpower is organized to complete this work, the inference model obtained by machine learning will be calculated in the general office computer of the general file utilization department. There is a certain delay. cause application inconvenience.

SUMMARY OF THE INVENTION

The purpose of the present invention is to provide a method for protecting private data in a file table image. It has fast calculation speed and strong anti-interference ability. It can accurately identify old files, forms with color seals, original alteration traces and extremely light table lines, and has good compatibility. Tables with different header styles can be effectively identified and processed.

The technical scheme of the present invention: a method for protecting privacy data in a file table image, which is characterized in that it specifically includes the following steps:

A method for protecting private data in an archive table image, specifically comprising the following steps:

1) First use the RGB color difference of the color image to dilute the content such as the color seal of the table image;

2) Scale the image to a specific size, design a horizontal edge detection operator with good compatibility under this image size, and obtain the edge intensity map of the image through the convolution operation;

3) Use Hough transform to obtain candidate horizontal and vertical line equations;

4) According to the straight line equation, carry out line tracking, use the circular queue to memorize the line coordinates of no more than 15 pixels, and judge the bending of the line in real time; for the excessive bending, it is judged as interference, and the coordinate information stored in the queue is used to perform a backtracking search other possible tracking trajectories;

5) Sort the Y coordinate of the horizontal line running through the table, and identify the header according to the feature that the header row height is quite different from the common row;

6) Using the graphical interactive interface, obtain the coordinate information of the utilization line from the file user, and combine the table information identified in the above steps to automatically blur the private part of the file image and retain the part that needs to be used.

The detection operator described in step 2) is to construct an edge detection operator matrix H with n rows and m columns. The sum of values is 1, and the sum of negative elements is -1, where n is an integer not greater than 7 and m is an integer not greater than 5.

Using Hough transform to obtain candidate horizontal line and vertical line equations described in step 3) includes the following two steps:

Perform Hough transform on the image, referred to as voting: take the parameter space of voting as a two-dimensional space: the row coordinates represent the straight line intercept, the column coordinates represent the straight line inclination angle, and the height is consistent with the height of the document; the voting threshold is the average of the edge strength reaching +5. There is 1 vote qualification to ensure that weak edges and strong edges, in the case of the same length, get the same number of votes, which is convenient to distinguish those table horizontal lines of the same length in the voting results; the edge strength is greater than 0 and lower than +5. , considered to be line artefacts produced by paper and scanning equipment;

Obtain the equation parameters of the candidate line: In the parameter space diagram of Hough change, the maximum value is obtained as the table line width maxValH; traverse the parameter space diagram, and all local maximum values that reach 70% of maxValH are investigated. In order to try to detect the intermittent missing table lines.

In step 4), the line segment tracking algorithm that can track appropriate bends, discontinuities and severe stroke interference is that, under the guidance of the straight line equation, the line is tracked, and the inclination angle and intercept of the straight line are obtained by the parametric equation of the straight line. The Y coordinate of the intersection of the straight line with X=0; in the edge intensity map, starting from the X=0, Y=intercept point, walk to the right at the inclination angle of the straight line; take t=20 as the threshold, and take the upper middle of the current point The point where the maximum value of the next three points is located is the tracking direction; when the intensity of the maximum point is greater than t, it is regarded as a line point, otherwise it is regarded as a non-line point; a circular queue is used to record the coordinates of the previously tracked points; if the line length is greater than 7, then calculate the bending angle of the line constructed by the last 7 points; when the angle is greater than 3°, it is regarded as a tracking error, and a rollback is performed; all tracked lines are recorded in the set S.

As described in step 5), the horizontal lines running through the table are sorted by Y coordinates, and the operation of identifying the header is to count the votes on the x coordinates of the left and right endpoints of all lines in S, and take the x with the largest number of votes at the left end as the left side of the table. Boundary L, the x with the largest number of votes at the right end is used as the right boundary R of the table; then traverse all the lines in S, and only keep the lines whose left and right endpoints are near L and R, which are regarded as horizontal lines that can run through the table; these horizontal lines are sorted by y-coordinate , treated as adjacent table horizontal lines. In the upper part of the table, judge from top to bottom: if the row height of two adjacent rows exceeds 20%, the previous row is regarded as the header part.

Step 6) The automatic blurring of the privacy part and the part to be used in the archive image refers to obtaining the row selected by the archive user, retaining a clear image for the row and the header, and performing conventional image blurring on other rows to obtain the final result. Available images for printout.

The beneficial effects of the invention are as follows: 1. The calculation speed is fast, and the identification and privacy blurring can be completed within 1 second on an ordinary office computer; Tables with very light lines can be accurately identified. 3. Good compatibility, it can effectively identify and process tables with different aspect ratios, different numbers of rows and columns, and different header styles.

Description of drawings

Figure 1 is a color scan (names have been masked) with various interference features;

Fig. 2 is the comparison of the grayscale method adopted by the ordinary color conversion to grayscale and this patent;

Figure 3 is a horizontal edge map;

Figure 4 shows the Hough transform parameter space (left) and the horizontal tracking effect (right);

Figure 5 is a horizontal line that is retained to get through the table;

6 is a schematic diagram of a straight line tracking algorithm;

Tracking from left to right, yellow for no bars, green for correct tracking, and red for interference lines

FIG. 7 is an image of protecting privacy information automatically generated after a user selects a utilization line in a graphical interactive interface.

Detailed ways

The present invention will be further described below in conjunction with the examples, but not as a basis for limiting the present invention.

Embodiment 1: A method for protecting private data in archive table images

Step 1: Uniform size and fade color stamps and signatures

Define the length and width of the original image as W and H, and the zoom ratio as zoom=min(1024/W, 1024/H). Scale the original image to zoom times.

In the image after size scaling, the line width can be controlled within 0.5 to 2 pixels, which is convenient for subsequent detection. The color image is separated into images of three channels, R, G, and B. Through the max calculation of the image matrix, Gray=max(R, G, B) is obtained, and Gray is the image after lightening the color, which can greatly reduce the interference of color seals, red signatures, etc. As shown in Figure 2 for comparison.

Step 2: Construct a horizontal edge detection operator to perform horizontal edge detection on the image

Construct the edge detection operator matrix H with n rows and m columns. It is recommended to take values 5 and 5 for n and m. The value of the middle row of the matrix is negative, and the value of the elements of both ends is positive. The value of each row is the same, and the value of all positive elements of the matrix is the same. The sum is 1, and the sum of negative element values is -1. The recommended value of H is:

H=[+0.100,+0.100,+0.100,+0.100,+0.100;

-0.025,-0.025,-0.025,-0.025,-0.025;

-0.150,-0.150,-0.150,-0.150,-0.150;

-0.025,-0.025,-0.025,-0.025,-0.025;

+0.100,+0.100,+0.100,+0.100,+0.100;]

Perform 2-dimensional convolution calculation on Gray and H to obtain the horizontal edge intensity map E=conv2d(Gray, H); the effect is shown in Figure 3.

The conv2d here and the 2d convolution of the convolutional neural network is a calculation.

Step 3: Extract table row lines

Data related to file privacy is generally in behavioral units, all of which are illustrated here with the tracking of horizontal lines.

vote. Perform Hough transform (voting) on the image, and take the parameter space of voting as a two-dimensional space: the row coordinates represent the intercept of the line, and the column coordinates represent the inclination angle of the line (with an angular resolution of 0.1° and an angle range of ±3°). There are a total of 61 columns. If necessary, you can expand this range by yourself, and the height is consistent with the height of the document. The voting threshold is that those whose edge strength reaches +5 are eligible for 1 vote, ensuring that weak edges and strong edges have the same number of votes in the case of the same length, which is convenient to distinguish those horizontal lines of the same length in the voting results. Edge strengths greater than 0 and lower than +5 are considered to be line artifacts produced by paper and scanners.

Get the equation parameters of the candidate line. In the parameter space diagram of Hough change (the black part on the left of Figure 4), the maximum value is obtained as the table line width maxValH. Traversing the parameter space graph, where the local maximum value of 70% of maxValH is reached, it is inspected in order to detect the discontinuous missing table lines as much as possible.

Step 4: Under the guidance of the straight line equation, trace the line (as shown in Figure 6). The parametric equation of the straight line can obtain the inclination angle and intercept of the straight line (the Y coordinate of the intersection with the straight line with X=0). In the edge intensity map, starting from the point of X=0, Y=intercept, walk to the right with a straight line inclination angle. Take t=20 as the threshold, and take the point where the maximum value of the upper, middle and lower points of the current point is located as the tracking direction. When the intensity of the maximum point is greater than t, it is regarded as a line point, otherwise it is regarded as a non-line point.

Use a circular queue (a data structure) to record the coordinates of previously tracked points. If the line length is greater than 7, calculate the line bending angle constructed by the nearest 7 points. When the included angle is greater than 3°, it is regarded as a tracking error and a rollback is performed. All traced lines are recorded in set S.

The reason for using the circular queue is: to examine the curve of the traced line, only the 7 points that have been traced recently are examined, and the coordinate storage space of the points in a longer range can be recycled.

Step 5: Obtain the horizontal through line and header of the table

Take the x-coordinates of the left and right endpoints of all lines in S to count votes, take the x with the largest number of votes at the left end as the left border L of the table, and the x with the largest number of votes at the right end as the right border R of the table.

Then traverse all the lines in S, and only keep the lines whose left and right endpoints are near L and R respectively, which are regarded as horizontal lines that can run through the table. The effect is shown in Figure 6.

Sort these horizontal lines by their y-coordinates and treat them as adjacent table horizontal lines. In the upper part of the table, judge from top to bottom: if the row height of two adjacent rows exceeds 20%, the previous row is regarded as the header part.

Step 6: Apply Privacy Shield

From the software interactive interface, obtain the row selected by the file user, keep a clear image of the row and the header, and perform conventional image blurring on other rows to obtain the final usable image for print output. The effect is shown in Figure 7.

Claims (6)

1. a method for protecting the privacy data in the archive table image, is characterized in that, specifically comprises the steps: 1) First use the RGB color difference of the color image to dilute the content such as the color seal of the table image; 2) Scale the image to a specific size, design a horizontal edge detection operator with good compatibility under this image size, and obtain the edge intensity map of the image through the convolution operation; 3) Use Hough transform to obtain candidate horizontal and vertical line equations; 4) According to the straight line equation, carry out line tracking, use the circular queue to memorize the line coordinates of no more than 15 pixels, and judge the bending of the line in real time; for the excessive bending, it is judged as interference, and the coordinate information stored in the queue is used to perform a backtracking search other possible tracking trajectories; 5) Sort the Y coordinate of the horizontal line running through the table, and identify the header according to the feature that the header row height is quite different from the common row; 6) Using the graphical interactive interface, obtain the coordinate information of the utilization line from the file user, and combine the table information identified in the above steps to automatically blur the private part of the file image and retain the part that needs to be used. 2. the method for protecting the privacy data in the archive table image according to claim 1, is characterized in that: the detection operator described in step 2) is, constructs the edge detection operator matrix H of n rows and m columns, matrix The value of the middle row is negative, the value of the row elements at both ends is positive, the value of each row is the same, the value of all positive elements of the matrix is 1, and the value of negative elements is -1, where n is an integer not greater than 7, and m is not greater than An integer of 5. 3. the method for the privacy data in the protection file form image according to claim 1, is characterized in that: utilize Hough transform described in step 3) to obtain candidate horizontal line and vertical line equation, comprise the following two steps: Perform Hough transform on the image, referred to as voting: take the parameter space of voting as a two-dimensional space: the row coordinates represent the straight line intercept, the column coordinates represent the straight line inclination angle, and the height is consistent with the height of the document; the voting threshold is the average of the edge strength reaching +5. There is 1 vote qualification to ensure that weak edges and strong edges, in the case of the same length, get the same number of votes, which is convenient to distinguish those table horizontal lines of the same length in the voting results; the edge strength is greater than 0 and lower than +5. , considered to be line artefacts produced by paper and scanning equipment; Obtain the equation parameters of the candidate line: In the parameter space diagram of Hough change, the maximum value is obtained as the table line width maxValH; traverse the parameter space diagram, and all local maximum values that reach 70% of maxValH are investigated. In order to try to detect the intermittent missing table lines. 4. the method for the privacy data in the protection file form image according to claim 1, is characterized in that: in step 4), the line segment tracking algorithm that can track suitable bend, discontinuity and severe stroke interference is, under the guidance of straight line equation , trace the line, obtain the inclination angle and intercept of the line through the parameter equation of the line, the intercept is the Y coordinate of the intersection with the line with X=0; in the edge intensity map, from X=0, Y=intercept point Start, walk to the right at the inclination angle of the straight line; take t=20 as the threshold, take the point where the maximum value of the upper, middle and lower points of the current point is located as the tracking direction; when the intensity of the maximum point is greater than t, it is regarded as a line point, Otherwise, it is considered as a non-line point; a circular queue is used to record the coordinates of the previously tracked point; if the line length is greater than 7, the bending angle of the line constructed by the latest 7 points is calculated; when the included angle is greater than 3°, it is regarded as a tracking error , go back; record all the traced lines into the set S. 5. the method for the privacy data in the protection file table image according to claim 1, is characterized in that: described in step 5) carries out Y coordinate sorting to the horizontal line that runs through table, the operation that identifies table header is, will The x-coordinates of the left and right endpoints of all lines in S are counted, and the x with the largest number of votes at the left end is taken as the left border L of the table, and the x with the largest number of votes at the right end is taken as the right border R of the table; then traverse all the lines in S, only keep the left and right endpoints Lines near L and R, respectively, are regarded as horizontal lines that can run through the table; these horizontal lines are sorted by y-coordinate and regarded as adjacent table horizontal lines. In the upper part of the table, judge from top to bottom: if the row height of two adjacent rows exceeds 20%, the previous row is regarded as the header part. 6. the method for protecting the privacy data in the archive form image according to claim 1, is characterized in that: the privacy part in the described automatic blurring archive image of step 6) and the part that retains need to utilize refer to, obtain archives and utilize For the row selected by the user, keep a clear image for the row and the header, and blur other rows with conventional images to obtain the final usable image for print output.

Images