Application of Image Processing Technology in License Plate Recognition

The automatic license plate recognition system can process the input car image through processing and output it as a license plate string of several bytes, which has unparalleled superiority in terms of storage space and connection with the management database. In the large parking lot, the traffic department's violation monitoring (electronic police), highways and bridge toll station management, etc., have broad application prospects.
1 License plate positioning and pre-processing After the car image file is input into the computer as a Raw format file, the computer extracts the license plate part from the entire image to realize the license plate location. Set the threshold to 127 and set the detection threshold to 16. Then, the image is scanned line by line from top to bottom. If the number of changes of 0→1 and 1→0 of a certain row is greater than the threshold, it is assumed to be the lowest point of the license plate to be tested, and the progressive scan is continued until 0→1 and 1→0. The case where the number of changes is less than 8 occurs. This value is assumed to be the highest point of the license plate to be tested. If the difference between the highest point and the lowest point is greater than 15, the target is considered to have been detected, otherwise the scanning is continued. If the target that meets the above conditions is not detected, the above threshold is automatically adjusted to repeat the above operation. Until you find the target.
Using the projection of the binary image in the vertical direction as a feature, the center point coordinates of the target are searched from left to right. Examine the previously obtained target height as the sum of the vertical projections in the square window of the side length (ie, the number of pixel points containing the pixel value of 1), if the value is less than the empirical threshold (multiple trials) If the threshold is taken as 150), it is regarded as the background part of the textless information. If the value is greater than the threshold for the first time, it is regarded as the left boundary part of the license plate to be recognized; after that, if the projection and the first time jump out of the cycle from large change, take the The midpoint abscissa of the window is the target center point. The number of pixels with a pixel value of 1 is again counted from the target center point to the right, and the height is the target height and the width is 30. If the value is less than the empirical threshold for the first time, it is considered to have reached the target right. The boundary, and take the coordinates of the point as the coordinates of the rightmost point of the target. The determination of the coordinates of the leftmost point of the target is equally acceptable.
Since the aspect ratio of the license plate is fixed, it is regarded as a target evaluation standard. Considering the deformation factor, if the aspect ratio is not within the interval (0.2 to 0.6), it is regarded as an invalid target. After the threshold is corrected, the cycle starts, and finally Reach the border.
Target image preprocessing includes image smoothing, separation of characters from the background, range adjustment, and tilt correction.
According to the actual situation, the smoothing of the image adopts the eight neighborhood averaging method. The mask used is:

This article refers to the address: http://

The thresholding algorithm used to implement character and background separation is to store the number of pixel points of all gray values ​​i in the target image in a one-dimensional integer array element A[i] having 256 elements. The comparison yields a gray value a having the greatest probability in the target image. The study found two different situations, as shown in Figures 1 and 2, respectively.

For case one, the image information is mainly located between the grayscale intervals (0 to a), and then the grayscale interval (0 to b) is found, so that the number of pixel points in the interval occupies the total pixel of the target image. 30% of the number. Let b be the threshold value, so that the pixel point whose gray value is greater than the threshold value takes a value of 0, and the pixel point of other cases takes a value of 1. Case 2 is handled in the same way. The target image is scanned from left to right with a detection window of the same height as the target image and having a width of 30. The pixel density is examined, and scanning is stopped when the number of pixels having a value of 1 is less than 50%. Take the left coordinate of the detection window at this time as the left boundary of the target. The right border of the target is equally available. Based on the range information of the obtained license plate image, the tilt correction is performed by rotation conversion if necessary.
2 Automatic single-character column-cutting and dividing is to divide the license plate image extracted after positioning into a single character image. The projection of the character block in the vertical direction necessarily takes a local minimum at the correct segmentation position (ie, at the gap within the character or character), and this position should satisfy the writing rules and character size limits. The vertical projection of the character image. The projection values ​​of the respective coordinates are detected from left to right in the horizontal direction. The coordinate that the first projection value is not 0 is detected as the left boundary of the first character, and the coordinate of the first projection value detected from the coordinate to the right is regarded as the right boundary of the first character, and the boundary coordinates of the remaining characters. The same is available.
The possible misclassification is removed by the average word width of the character and the average distance between the two-character left border. A character whose word width is smaller than the average word width by a certain ratio (such as 0.2) is regarded as an invalid character; if the distance between the two characters is less than the average distance and the sum of the distance and the word width is not greater than the average distance, the combination is one character; A certain ratio of the average word width (such as 2.4) is considered to be adhesion of two characters.
After the above processing, accurate segmentation results can be obtained. The characters are transformed into a 64×64 dot matrix space to facilitate the processing of subsequent feature extraction and other stages.
3 Contouring and refinement contouring uses the four-neighbor method to scan the 64×64 text image f(i,j) after noise smoothing, scanning the upper, lower and left sides of the black pixel point (i,j). The right four neighbors, as long as there is a point that is not black, the point (i, j) is the point on the outline of the character, and the gray level is set to 1 (ie, black). In other cases, the gray of the point (i, j) is made. Degree is 0 (ie white). The refinement process uses the secondary scan refinement method, which is faster, but because it is a relatively simple iterative algorithm, it sometimes causes a certain degree of skeleton deformation.
Figures 3(b) and 3(c) show the results of the contouring and refinement treatments, respectively.

4 Extraction of microstructural features divides the characters into n×n networks, and the directional characteristics of regional strokes are counted for each small network. Each small area highlights the local features of the character and is insensitive to minor offsets or distortions. The two lines formed by the adjacent three points are defined as microstructures. The contoured characters, where there are twelve boundary line elements, are related to the character strokes (as shown in Figure 4).

According to the four basic directions of the character stroke, the horizontal, vertical, and ±45° four line directions can be defined accordingly. And the regional stroke direction density vector can be counted.
The 64×64 characters to be recognized are divided into 5×5 grids. The size of the first 4×4 grid is 13×13, and the last row and column grid are 12×12 except for the last one. The last behavior is 12×. 13, the last column is 13×12, and the eigenvectors of the stroke direction are counted, so that a horizontal, vertical, +45°, -45° four-dimensional directional feature is obtained in each region, which constitutes the 100-dimensional classification feature of the whole character. .
The stability of the extracted features is crucial to the correct rate of recognition, so the characters are subjected to double partitioning of 8×8 and 7×7 in the fine classification. The regional stroke direction vectors (four directions) of 64+49=113 small areas are respectively counted to form a fine classification feature of 113×4=452 dimensions. The 8x8 segmentation is used to extract finer structural features in a smaller area. In order to prevent the instability of the segmentation edge, a 7×7 double segmentation is performed, so that the stroke of the most unstable 8×8 mesh edge is in the central stable region of the 7×7 grid, which improves the edge of the region. stability.
5 Matching strategy In order to improve the accuracy and speed of recognition, a multi-level classification identification scheme is adopted in the matching.
In the rough classification, the simple regional stroke direction feature is used to divide the characters into 5×5 grids (25 small areas in total) to calculate the four directional characteristics of the line elements, and form a 100-dimensional (25×4) feature vector. The absolute distance distance criterion is adopted. Let any feature vector in the dictionary library be Pji=(pj1, pj2, pj3, pj4, Λ, pj100), and the feature vector of the character to be recognized is h=(h1, h2, h3, h4, Λ, h100), dictionary The distance between any one of the templates and the character to be recognized is dj.

In the dj, the first 10 characters with the smallest value are selected as the result of the preliminary matching, and the next step is to perform fine classification.
In the fine classification, the candidate characters are extracted by the double segmentation into 452-dimensional feature vectors as features of the fine classification. The second match is made using a discriminant criterion similar to the coarse classification. The parameters are determined through experiments, and the different weight coefficients are combined with the coarse classification criteria to determine the matching degree between the characters to be recognized and the different standard templates. The first four are taken as the final result and output to the specified text file.
6 The establishment of the standard dictionary library is preferred among many fonts. The Chinese characters are selected from the Song font library, and the letters and numbers are selected from the OCR-A font. The standard characters are normalized, contoured, and feature extracted separately. The standard template is the feature vector from which the feature is extracted.
7 test results license plate positioning is very ideal; character segmentation is correct; the recognition of the first character of Chinese characters sometimes misunderstands (visible Chinese characters are difficult to identify, the matching algorithm and template library is the key to the problem); The number is better recognized; it reaches 100% in the first two levels of the fine classification priority.

Yacenter has experienced QC to check the products in each process, from developing samples to bulk, to make sure the best quality of goods. Timely communication with customers is so important during our cooperation.

If you can't find the exact product you need in the pictures,please don't go away.Just contact me freely or send your sample and drawing to us.We will reply you as soon as possible.

Obd Harness

Obd Harness,Obd2 Wiring Harness,Obd1 Harness,Obd Connector Wire Harness

Dongguan YAC Electric Co,. LTD. , http://www.yacenter-cn.com