CN104318233A - Method for horizontal tilt correction of number plate image - Google Patents

Abstract

The invention discloses a method for correcting the horizontal inclination of a license plate image. Due to the shooting angle and other reasons of the license plate image, the license plate will be tilted, which is not conducive to character segmentation, so the license plate needs to be corrected. After a series of image enhancement operations, the edge image quality of the obtained license plate is better. Based on the edge image, the present invention uses the rules of characters to find the legal character connected domain of the edge image, and performs linear fitting on the upper and lower endpoints of the connected domain to find the characters. The upper and lower boundaries of the area are calculated to obtain the slope, and then the rotation angle of the tilt correction is calculated to rotate the license plate. The invention improves the removal method of the pseudo-connected domain and the search method of the connected domain, and greatly improves the robustness of the tilt correction algorithm.

Description

Horizontal tilt correction method for license plate image

technical field

The invention relates to a method realized based on character connected domains in a license plate image, in particular to a method for correcting the horizontal inclination of the license plate image.

Background technique

In intelligent transportation system (ITS, Intelligent transportation system), automatic license plate recognition (LPR, License Plate Recognition) plays an important role. Generally, LPR consists of the following steps: image acquisition, license plate location, license plate character segmentation, and recognition. However, due to the shooting angle and other reasons, the license plate in the image will appear tilted, as shown in Figure 1. This will cause problems such as connection and destruction between license plate characters, which is not conducive to character segmentation and recognition. Therefore, the importance of license plate image tilt correction is obvious.

Some existing correction methods generally obtain the border line of the license plate, and then calculate the inclination angle. However, the license plate has its complexity. On the one hand, the frame of the license plate may not be a standard straight line. On the other hand, the image obtained by the positioning algorithm does not have a frame. In addition, there are some correction algorithms, either the calculation is time-consuming and the correction rate is low, or they are greatly disturbed by the image background. In view of this, the present invention starts from the characteristics of the characters themselves, that is, the characters are arranged neatly in the license plate, and the upper and lower endpoints are respectively on two straight lines. Optimizing the skew correction algorithm based on character connected domains, mainly improving the removal method of pseudo-connected domains and the search method of connected domains, so that the robustness of the skew correction algorithm is improved, and it has been successfully applied to license plates in many states in the United States Identifying.

Contents of the invention

In view of the above problems, the present invention proposes a method for correcting the horizontal inclination of the license plate image.

Concrete scheme of the present invention is:

A method for correcting the horizontal inclination of a license plate image comprises the following steps:

1) Find all connected domains on the edge image of the license plate, the connected domains contain position information, and map the connected domains to the original license plate image;

2) Starting from the characteristics of the character itself, search for credible character connected domains from the connected domains found in step 1). The method of searching for credible character connected domains is to first search for credible character connected domains based on the edge image. When based on When the number of trusted character-connected domains found in the edge image is greater than or equal to 3, proceed to the next step; when the number of trusted character-connected domains found based on the edge image is less than 3, turn to finding credible character-connected domains based on the binary image area;

3) Fitting the upper and lower boundary points into a straight line using the least squares method, and selecting the accurate upper and lower boundary points of the character area;

4) Obtain the slopes of the straight lines at the upper and lower boundaries of the character area, obtain the rotation angle, and calculate the positions of the upper and lower boundaries of the character area after rotation according to the rotation angle.

The characteristic of the characters is that the characters are neatly arranged in the license plate, and the upper and lower endpoints are respectively on two straight lines.

The credible character connected domain based on the edge image search is specifically:

a) judging whether the height, horizontal position and spacing of two connected domains are less than a set threshold, and merging the two connected domains which are less than the set threshold;

b) judging the character area and character aspect ratio of the connected domain, and deleting the connected domain whose character area and character height-width ratio do not belong to the set range;

c) According to the characteristic that the license plate characters are concentrated in a certain height range, delete the isolated connected domain;

d) According to the characteristic that the character height of the license plate itself is basically consistent, the average height of the connected domain is obtained through cluster analysis. If the difference between the height of a connected domain and the average value is greater than the set threshold, the connected domain is deleted. Otherwise, keep the connected domain;

e) Calculate the absolute value of the difference between these slopes according to the slope of the straight line formed by the central points of the two connected domains, and delete the connected domains that do not meet the conditions at both ends according to the absolute value of the difference;

f) According to the distance between the connected domains, the connected domains greater than 3 times the average width of the connected domains are regarded as outlier connected domains, and the outlier connected domains are deleted.

The step c) is specifically as follows: set a straight line L, scan the edge image of the license plate from top to bottom, find the straight line L', make it pass through the largest number of connected domains, and delete the upper boundary below L' or the lower boundary Connected domains above L'.

Described step e) is specifically: the slope of the straight line formed by the center of the first and the penultimate connected domain is a, the slope of the straight line formed by the second and the center point of the penultimate connected domain is b, and the calculation of abs( a-b) If it is less than the set threshold, it is considered to meet the condition; if it is greater than the set threshold, it is considered that there is a connected domain at both ends that needs to be deleted, and then calculate the slope c of the straight line formed by the center point of the first and last connected domain, and compare The size of abs(a-c) and abs(b-c), if abs(a-c)>abs(b-c), delete the penultimate connected domain, otherwise, delete the first connected domain; repeat step e for the remaining connected domains ) until the conditions are met.

The method for finding credible character connected domains based on binary images is specifically:

A) Find all connected domains in the binary image;

B) judging whether the height, horizontal position and spacing of two connected domains are less than a set threshold, and merging the two connected domains smaller than the set threshold;

C) judging the character area and character aspect ratio of the connected domain, and deleting the connected domain whose character area and character height-width ratio do not belong to the set range;

D) According to the feature that the characters of the license plate are concentrated in a certain height range, set a straight line H, scan the binary image of the license plate from top to bottom, find the straight line H', and make it pass through the largest number of connected domains. Delete the connected domain whose upper boundary is below H' or whose lower boundary is above H';

E) According to the characteristic that the character height of the license plate itself is basically consistent, the average height of the connected domain is obtained through cluster analysis. If the difference between the height of the connected domain and the average value is greater than the set threshold, the connected domain is deleted, otherwise it is retained. the connected domain;

F) Calculate the absolute value of the difference between these slopes according to the slope of the straight line formed by the central points of the connected domains, and delete the connected domains at both ends that do not meet the conditions according to the absolute value of the difference;

G) According to the distance between the connected domains, the connected domains greater than 3 times the average width of the connected domains are regarded as outlier connected domains, and the outlier connected domains are deleted.

H) If the number of credible connected domains of characters found in the binary image satisfies the condition, proceed to the subsequent steps; otherwise, it is judged that the license plate image does not contain license plate information, and the correction method is ended.

The step F) is specifically: the slope a of the straight line formed by the center of the first and the penultimate connected domain, the slope b of the straight line formed by the second and the center point of the penultimate connected domain, and calculate abs(a-b) If it is less than the set threshold, it is considered to meet the condition; if it is greater than the set threshold, it is considered that there is a connected domain at both ends that needs to be deleted, and then calculate the slope c of the straight line formed by the center points of the first and last connected domains, and compare abs( The size of a-c) and abs(b-c), if abs(a-c)>abs(b-c), then delete the last connected domain, otherwise, delete the first connected domain; repeat the judgment of step F) for the remaining connected domains, until the conditions are met.

Described step 3) specifically is: find the highest point and the lowest point of each connected domain, these points are regarded as the upper and lower boundary points of the connected domain, carry out linear fitting to all upper and lower boundary points respectively with the least squares method, judge the upper and lower boundaries Among the points, whether the distance between the point farthest from the respective fitting straight line and the fitting straight line is less than or equal to the threshold value, if it is less than or equal to the threshold value, continue to the next step, if it is greater than the threshold value, delete the farthest point, and repeat the remaining points Carry out linear fitting by the least squares method until the distance between the farthest point after fitting and the fitting line is less than or equal to the threshold, and finally obtain the upper and lower boundary lines.

Finding the angle of rotation in the described step 4) is specifically: if the absolute value of the difference of the slope of the upper and lower boundary straight lines is less than the set threshold, then the slopes of the upper and lower boundary straight lines are added and averaged to obtain the rotation angle of the license plate; Otherwise, select the slope of the fitting line with the largest number of reserved boundary points to calculate the rotation angle of the license plate.

The position of the upper and lower boundaries of the character area after the rotation is calculated according to the rotation angle is specifically:

The calculation method of the upper boundary position is to select the point located at the upper end of the upper boundary line and the closest to the fitting line, and calculate the position of this point in the rotated image according to the obtained rotation angle;

The calculation method of the lower boundary position is to select the point located at the lower end of the lower boundary line and the closest to the fitting line, and calculate the position of this point in the rotated image according to the obtained rotation angle;

Its calculation formula is as follows:

x=Xcosα-Ysinα

y=Xsinα+Ycosα

Where (X, Y) is the coordinates of the point closest to the fitting line before rotation, (x, y) is the coordinates of the point closest to the fitting line after rotation, and α is the rotation angle.

The innovation point of the present invention lies in starting from the characteristics of the characters themselves, that is, the characters are neatly arranged in the license plate, and the upper and lower endpoints are respectively on two straight lines. By finding the upper and lower boundary points of the connected domain of each character, the upper and lower boundary points are fitted into two straight lines using the least square method, and the slope of the straight line can reflect the angle of the license plate tilt. In addition, the invention improves the removal method of the pseudo-connected domain and the search method of the connected domain, greatly improves the robustness of the tilt correction algorithm, and is successfully applied to the license plate recognition of many states in the United States.

Description of drawings

Figure 1(a) is the model when the horizontal inclination angle of the license plate image is α>0;

Figure 1(b) is the model when the horizontal inclination angle of the license plate image is α<0;

Figure 2(a) is the model when the license plate image has a vertical tilt angle θ>0;

Figure 2(b) is the model when the license plate image is tilted vertically at an angle θ<0;

Figure 3 shows the top boundary point and the bottom boundary point of the character area in the license plate;

Figure 4 shows that the least squares method is used to carry out straight line fitting to the top and bottom boundary points of the characters in the license plate;

Figure 5(a) is the license plate grayscale image G1;

Figure 5(b) is the license plate image after histogram equalization;

Fig. 5 (c) is the license plate image after Gaussian filter smoothing;

Figure 5(d) is the license plate image G2 sharpened by Laplace;

Figure 5(e) license plate binarized image B1;

Figure 5(f) is the Canny edge detection image C1 of the license plate;

Figure 6(a) maps the connected domain to the original image;

Figure 6(b) The connected domains within the character "X" are merged;

Figure 6(c) The illegal connected domain is deleted;

Figure 6(d) shows that the connected domain in the character "8" is deleted;

Figure 6(e) shows that the connected domains in the characters "H" and "2" are deleted;

Figure 6(f) deletes the license plate image before the last connected domain;

Figure 6(g) The license plate image after deleting the last connected domain;

Figure 6(h) deletes the license plate image before the rightmost outlier connected domain;

Figure 6(i) License plate image after removing the rightmost outlier connected domain;

Figure 7(a) is the license plate edge image; Figure 7(b) is the connected domain detected based on the edge image mapped to the original image;

Figure 7(c) maps the detected two legitimate connected domains on the original image;

Figure 7(d) binary image of the license plate;

Figure 7(e) is the mapping of the connected domain detected based on the binary image to the original image;

Figure 7(f) maps the detected legal connected domains to the original image;

Figure 8(a) is the license plate image after linear fitting of the highest point and the lowest point of each connected domain with the least square method;

Figure 8(b) is the license plate image after deleting the upper and lower points of the character "6";

Fig. 9 (a) is the license plate image after linear fitting is carried out to the remaining upper and lower boundary points respectively with the least squares method;

Figure 9(b) is the image after horizontal tilt correction of the license plate;

Figure 10 is a flow chart of the image horizontal tilt correction algorithm.

Detailed ways

There are two situations for the tilt direction of the license plate area: horizontal tilt, as shown in Figure 1; vertical tilt, as shown in Figure 2. In Figure 1, only the angle α between the axis X' of the inclined area and the horizontal axis X is unknown. Once the angle α is obtained, the entire license plate image can be corrected by only rotating -α. In Figure 2, only the angle θ in the vertical direction is unknown. In practical applications, there are two kinds of inclinations in most license plates at the same time. Therefore, the inclination in the horizontal direction is corrected first, and then the inclination in the vertical direction is corrected. The algorithm in the present invention only involves corrections in the horizontal direction.

The colors of the characters in the license plate are the same or close, so in the binary image of the license plate, a complete independent character will become a connected domain. On the other hand, due to the relatively high contrast between the character color and the background color, the edge of the character can be obtained by an edge extraction algorithm, such as the Canny algorithm. A complete character edge is also a connected domain. Some strategies can be exploited to keep only the connected domains of characters. As mentioned above, in the license plate image, the top boundary points and bottom boundary points (called feature points) of the character area can reflect the inclination trend of the character area, as shown in Figure 3. After obtaining the top boundary points and bottom boundary points of these character areas, the least squares method is used to fit the top boundary points and bottom boundary points respectively, so that the inclination angle in the horizontal direction of the license plate can be obtained, as shown in Figure 4. Therefore, license plate tilt angle information can be obtained from the distribution trend of these points.

Taking the processing of the license plate grayscale image G1 (see Figure 5(a)) as an example, the specific implementation is described. Before tilt correction, some preprocessing operations are required on the license plate grayscale image. Preprocessing operations include histogram equalization (see Fig. 5(b)), filtering with a Gaussian smoothing filter (size 5x5, δ = 1.1) (see Fig. 5(c)), followed by second-order differential-based Laplac Adams sharpening operation (see Figure 5(d)). After the three-step preprocessing operation, a license plate grayscale image G2 is obtained. It is observed that the grayscale image G2 is more or less resistant to noise, and the edge information in it has also been enhanced. Finally, a binarized image B1 (see Figure 5(e)) and a Canny edge detection image C1 (see Figure 5(f)) are obtained. In B1, the characters are white and the background is black. Also, in C1, the edges are white and the background is black.

The specific steps of license plate image horizontal tilt correction are as follows:

(1) Find all connected domains on the edge image of the license plate. The connected domains contain position information, and map the connected domains to the original image, as shown in Figure 6(a).

(2) Searching for credible character-connected domains based on edge images mainly includes the following steps:

1) Merge separated connected domains of characters, because some characters such as "M", "X", and "H" may appear in the middle

The inter-connected domain is divided into two parts. In order to retain as many connected domains as possible, it is necessary to merge these possibly separated character connected domains. The principle of merging is to judge whether the two connected domains meet the requirements of similar height, upper boundary at the same horizontal position and spacing less than a certain threshold. As shown in Figure 6(b), the connected domains within the character “X” are merged.

2) Delete the illegal connected domains, delete the connected domains that do not meet the conditions through some prior knowledge, such as character area and character width-to-height ratio, and get Figure 6(c), it can be seen that most of the illegal connected domains have been deleted.

3) Delete the isolated connected domain, the distribution of license plate characters has certain characteristics, that is, the license plate characters are concentrated in a certain height range, so as shown in Figure 6(c), set a red line L and scan from top to bottom The image finds the position with the largest number of connected domains passing through L, and deletes those connected domains whose upper boundary is below L or whose lower boundary is above L. As shown in Figure 6(d), the connected domains in the character "8" are deleted.

4) Delete the connected domains with illegal heights, the height of the license plate characters is basically consistent, and obtain the average height avgHeight of the connected domains through cluster analysis, delete the connected domains with a large difference from avgHeight, as shown in Figure 6(e), the character Connected domains within "H", "2" are deleted.

5) Delete the connected domains that do not meet the conditions at both ends, the condition is the slope a of the straight line formed by the center of the first and the penultimate connected domain, the slope b of the straight line formed by the second and the center point of the penultimate connected domain, If the calculation abs(a-b) is greater than a certain threshold, it is considered that there is a connected domain at both ends that needs to be deleted, and then calculate the slope c of the straight line formed by the center points of the first and last connected domains, and compare abs(a-c) and abs(b-c) The size of , if abs(a-c)>abs(b-c), delete the last connected domain; otherwise, delete the first connected domain. As shown in Figure 6(f), the last connected domain is deleted, and Figure 6(g) is obtained.

6) Delete the connected domain of the outlier. By judging the distance of the connected domain, if the average width avgWidth of the connected domain is greater than 3 times, it means that it is an outlier point. As shown in Figure 6(h), there is an outlier connected domain on the far right, delete Figure 6(i) is obtained.

(3) Find credible character-connected domains based on binary images. In some cases, the effect of Canny edge detection is not good, as shown in Figure 7(a), (b), which may result in the finding of a small number of legitimate connected domains, or even no, as shown in Figure 7(c) As shown, there are only two legal connected domains, and skew correction cannot be performed. At this time, we can turn to find the connected domain of the binary image. Therefore, when the number of legal connected domains of the Canny edge image is less than 3, directly turn to the method of finding trusted connected domains of characters described in step (2) to find trusted connected domains in the binary image, and adaptive Adjust the binarization threshold so that the multiples of the threshold are {0.3, 0.5, 0.6, 0.8, 1.5, 1.7, 1.9, 2.1}, until the number of legitimate connected domains found in the binary image If the conditions are met, proceed to the subsequent steps; otherwise, it is judged that the image does not contain license plate information, and the correction algorithm is ended. Figure 7(d) is a binary image, all connected domains are shown in Figure 7(e), and there are 3 remaining legal connected domains, which meet the conditions, as shown in Figure 7(f).

(4) Calculate the rotation angle of the tilt correction, and find the upper and lower boundaries of the character area. As shown in Figure 8(a), find the highest point and the lowest point of each connected domain, use the least square method to perform linear fitting on all the upper and lower points, and delete the point farthest from the fitting line, Until the farthest distance is less than a certain threshold, stop deleting. As shown in FIG. 8(b), the dots at the upper and lower ends of the character "6" are deleted. The search method of the upper and lower boundary points of the connected domain is as follows:

2) Assume that the first character connected domain is expressed as Rs(x0, y0, w0, h0), where (x0, y0) represents the coordinates of the upper left point of the connected domain Rs, and (w0, h0) represents the width and sum of the connected domain Rs High; suppose the last connected domain is expressed as Re(x1, y1, w1, h1), where (x1, y1) indicates the coordinates of the upper left point of the connected domain Re, and (w1, h1) indicates the width and height of the connected domain Re. Select a point P0(x,y) on the left boundary of the connected domain Rs, where x=x0, y=y0+h0*0.1, and select a point P1(x',y') on the left boundary of the connected domain Re, where x '=x1,y'=y1+h1*0.1. Therefore, an upper boundary search line Lu can be obtained from these two points, as shown in Figure 8(a). Similarly, select a point P2(x2,y2) on the left boundary of the connected domain Rs, where x2=x0,y2=y0+h1*0,9; and select a point P3(x2',y2 on the left boundary of the connected domain Re '), where x2'=x0, y2'=y0+h1*0.9. Therefore, a lower boundary search Lb can be obtained from these two points, as shown in Figure 8(a).

2) Track each white point on Lu and Lb separately. Assuming point p is a white point on Lu, searching upwards can get the highest point ph in the same connected domain as point p. Therefore, the highest point of each connected domain can be obtained by Lu search. All these points can form an upper boundary point set, denoted as Sup, as shown in Fig. 8(b). Similarly, the lowest point of each connected domain can be obtained by searching downward through Lb, and all these points form the lower boundary point set, denoted as Sbtm, see Figure 8(b).

3) Delete the points that do not meet the conditions in the point sets Sup and Sbtm. As shown in Fig. 8(b), some points do not belong to the connected domain of arbitrary characters, but satisfy the above conditions of boundary points. In addition, character breaks will also affect the search for boundary points. In order to ensure that the slope of the boundary line can accurately reflect the slope angle information of the license plate, these points that do not belong to the connected domain of any characters must be deleted. The specific steps to delete these points are as follows: i) According to the least square method, use the straight line L1 to fit these points; ii) Calculate the average and variance of the distance from these points to the straight line L1; iii) When the average value of the distance is greater than a certain threshold , delete those points with the largest distance to L1; iv) return to step i), until the average value is less than a certain threshold. Therefore, the remaining points are eligible upper and lower boundary points.

4) Use the least square method to fit the remaining points in Sup and Sbtm into a straight line, and select the accurate upper and lower boundary points of the character area. Fit the upper and lower boundary points into straight lines L1 and L2, respectively, as shown in Figure 9(a). The slopes of the straight lines of L1 and L2 can be calculated, expressed as ku and kb. If the absolute value of the difference between the upper and lower slopes is less than a certain threshold, then add the upper and lower slopes and take the average value to obtain the rotation angle of the license plate. to calculate the rotation angle of the license plate. While obtaining the rotation angle, it is also necessary to obtain the upper and lower boundaries of the license plate character area. The method of obtaining the upper boundary is to select the point that is located at the upper end of the fitting line and the point closest to the fitting line, and then calculate the rotation angle of this point according to the obtained rotation angle. The coordinates in the image, this value is the value of the upper boundary, and the method of obtaining the lower boundary is to select the point at the lower end of the fitting line and the closest point to the fitting line, and then calculate the position of this point in the rotated image according to the obtained rotation angle The coordinates of , this value is the value of the lower boundary. Its calculation formula is as follows:

x=Xcosα-Ysinα

y=Xsinα+Ycosα

Where (X, Y) is the coordinates before the boundary point rotation, (x, y) is the coordinates after the boundary point rotation, and α is the rotation angle.

(5) After obtaining the rotation angle, the license plate is rotated according to this angle, and the upper and lower boundaries of the character area are also obtained, as shown in Figure 9(b).

Claims (10)

1. a license plate image horizontal tilt correction method, is characterized in that, comprises the steps: 1) Find all connected domains on the edge image of the license plate, the connected domains contain position information, and map the connected domains to the original license plate image; 2) Starting from the characteristics of the character itself, search for credible character connected domains from the connected domains found in step 1). The method of searching for credible character connected domains is to first search for credible character connected domains based on the edge image. When based on When the number of trusted character-connected domains found in the edge image is greater than or equal to 3, proceed to the next step; when the number of trusted character-connected domains found based on the edge image is less than 3, turn to finding credible character-connected domains based on the binary image area; 3) Fitting the upper and lower boundary points into a straight line using the least squares method, and selecting the accurate upper and lower boundary points of the character area; 4) Obtain the slopes of the straight lines at the upper and lower boundaries of the character area, obtain the rotation angle, and calculate the positions of the upper and lower boundaries of the character area after rotation according to the rotation angle. 2. The method for correcting the horizontal inclination of the license plate image as claimed in claim 1, wherein the character characteristic is that the characters are arranged neatly in the license plate, and the upper and lower endpoints are respectively on two straight lines. 3. The license plate image horizontal inclination correction method as claimed in claim 1, is characterized in that described based on edge image search credible character connected domain is specifically: a) judging whether the height, horizontal position and spacing of two connected domains are less than a set threshold, and merging the two connected domains which are less than the set threshold; b) judging the character area and character aspect ratio of the connected domain, and deleting the connected domain whose character area and character height-width ratio do not belong to the set range; c) According to the characteristic that the license plate characters are concentrated in a certain height range, delete the isolated connected domain; d) According to the characteristic that the character height of the license plate itself is basically consistent, the average height of the connected domain is obtained through cluster analysis. If the difference between the height of a connected domain and the average value is greater than the set threshold, the connected domain is deleted. Otherwise, keep the connected domain; e) Calculate the absolute value of the difference between these slopes according to the slope of the straight line formed by the central points of the two connected domains, and delete the connected domains that do not meet the conditions at both ends according to the absolute value of the difference; f) According to the distance between the connected domains, the connected domains greater than 3 times the average width of the connected domains are regarded as outlier connected domains, and the outlier connected domains are deleted. 4. The method for correcting the horizontal inclination of the license plate image as claimed in claim 3, wherein said step c) is specifically: setting a straight line L, scanning the edge image of the license plate from top to bottom, finding the straight line L', making it The number of connected domains passing through is the largest, and the connected domains whose upper boundary is below L' or whose lower boundary is above L' are deleted. 5. The horizontal inclination correction method of license plate image as claimed in claim 3, characterized in that said step e) is specifically: the slope of the straight line formed by the center of the first and the penultimate connected domain is a, and the slope of the second one is a. The slope of the straight line formed with the center point of the penultimate connected domain is b. If the calculated abs(a-b) is less than the set threshold, the condition is considered satisfied; if it is greater than the set threshold, it is considered that there is a connected domain at both ends that needs to be deleted. Then calculate the slope c of the straight line formed by the center points of the first and last connected domains, compare the size of abs(a-c) and abs(b-c), if abs(a-c)>abs(b-c), delete the last one connected Otherwise, delete the first connected domain; repeat the judgment of step e) for the remaining connected domains until the condition is met. 6. The license plate image horizontal inclination correction method as claimed in claim 2, characterized in that the method for finding credible character connected domains based on binary images is specifically: A) Find all connected domains in the binary image; B) judging whether the height, horizontal position and spacing of two connected domains are less than a set threshold, and merging the two connected domains smaller than the set threshold; C) judging the character area and character aspect ratio of the connected domain, and deleting the connected domain whose character area and character height-width ratio do not belong to the set range; D) According to the feature that the characters of the license plate are concentrated in a certain height range, set a straight line H, scan the binary image of the license plate from top to bottom, find the straight line H', and make it pass through the largest number of connected domains. Delete the connected domain whose upper boundary is below H' or whose lower boundary is above H'; E) According to the characteristic that the character height of the license plate itself is basically consistent, the average height of the connected domain is obtained through cluster analysis. If the difference between the height of the connected domain and the average value is greater than the set threshold, the connected domain is deleted, otherwise it is retained. the connected domain; F) Calculate the absolute value of the difference between these slopes according to the slope of the straight line formed by the central points of the connected domains, and delete the connected domains at both ends that do not meet the conditions according to the absolute value of the difference; G) According to the distance of the connected domain, the connected domain greater than 3 times the average width of the connected domain is regarded as the connected domain of the outlier, and the connected domain of the outlier is deleted; H) If the number of credible connected domains of characters found in the binary image satisfies the condition, proceed to the subsequent steps; otherwise, it is judged that the license plate image does not contain license plate information, and the correction method is ended. 7. The license plate image horizontal inclination correction method as claimed in claim 6, characterized in that said step F) is specifically: the slope a of the straight line formed by the center of the first and the penultimate connected domain, the second and The slope b of the straight line formed by the center point of the penultimate connected domain, if the calculated abs(a-b) is less than the set threshold, it is considered to meet the condition; if it is greater than the set threshold, it is considered that there is a connected domain at both ends that needs to be deleted, and then calculate The slope c of the straight line formed by the center points of the first and last connected domain, compare the size of abs(a-c) and abs(b-c), if abs(a-c)>abs(b-c), delete the last connected domain, otherwise, Delete the first connected domain; repeat the judgment of step F) for the remaining connected domains until the condition is met. 8. The license plate image horizontal tilt correction method as claimed in claim 1, characterized in that said step 3) is specifically: find the highest point and the lowest point of each connected domain, these points are regarded as the upper and lower boundary points of the connected domain , use the least squares method to linearly fit all the upper and lower boundary points, and judge whether the distance between the upper and lower boundary points and the farthest point from the respective fitting straight line is less than or equal to the threshold value, if it is less than or equal to the threshold value, continue to the next step , if it is greater than the threshold, delete the farthest point, and perform least squares linear fitting on the remaining points again until the distance between the farthest point after fitting and the fitting line is less than or equal to the threshold, and finally get the upper and lower boundaries straight line. 9. The license plate image horizontal inclination correction method as claimed in claim 1, is characterized in that in the described step 4), finding the angle of rotation is specifically: if the absolute value of the difference of the slope of the upper and lower boundary straight lines is less than the set threshold, Then add the slopes of the upper and lower boundary lines and take the average value to obtain the rotation angle of the license plate; otherwise, select the slope of the fitting line with the largest number of boundary points to calculate the rotation angle of the license plate. 10. The method for correcting the horizontal inclination of the license plate image as claimed in claim 1, wherein the position of the upper and lower boundaries of the character area after the rotation is calculated according to the rotation angle is specifically: The calculation method of the upper boundary position is to select the point located at the upper end of the upper boundary line and the closest to the fitting line, and calculate the position of this point in the rotated image according to the obtained rotation angle; The calculation method of the lower boundary position is to select the point located at the lower end of the lower boundary line and the closest to the fitting line, and calculate the position of this point in the rotated image according to the obtained rotation angle; Its calculation formula is as follows: x=Xcosα-Ysinα y=Xsinα+Ycosα Where (X, Y) is the coordinates of the point closest to the fitting line before rotation, (x, y) is the coordinates of the point closest to the fitting line after rotation, and α is the rotation angle.