Automatic license plate recognition system under complex background With the development of modern transportation, automatic license plate recognition technology has become an important part of intelligent transportation. The license plate recognition technology mainly uses computer image processing technology to analyze the image of the license plate to automatically extract the license plate information and determine the license plate number. Generally speaking, in the automatic license plate recognition system, the key technical issues to be dealt with are license plate location and character segmentation. Many methods have been proposed for the automatic license plate recognition system, such as the license plate location algorithm using multiple features, the license plate location method based on color and texture analysis, and the neural network algorithm for automatic license plate recognition. In view of the possibility of blurring and noise interference in the image dynamically collected by the camera, we first use an improved fuzzy C-means clustering algorithm to segment the collected image, and then locate the license plate according to the characteristics of the license plate area. After the license plate is located, the characters are segmented and recognized according to the distribution characteristics of the characters in the license plate. After experimenting with the images of the complex background collected, an ideal automatic license plate recognition effect is obtained. In order to improve the convergence rate of clustering, the membership degree needs to be corrected. A suppression threshold parameter β is introduced in the semi-suppressive fuzzy C-means clustering algorithm (HSFCM), and the maximum membership value uRj is compared with the threshold If it is greater than the threshold, it will be revised; otherwise, it will not be revised. In order to better improve the speed of clustering, the membership correction formula is changed to: (5) If || V (k) -V (k + 1) || <ε, stop, otherwise let k = k + 1 and repeat steps (2), (3), (4), (5) . 1.2 The positioning of the license plate According to the characteristics of the license plate image captured, the license plate is generally below the image, and most of the ground is below the license plate. In the horizontal direction, the grayscale distribution of the ground image is relatively uniform; and the license plate area is due to image characters The distribution of the image makes the change frequency of the image gray value in the horizontal direction is relatively large, and the change interval is relatively uniform. According to the above analysis, positioning the license plate from bottom to top can avoid the interference of the complex background above and shorten the positioning time. The changes in the horizontal gray values ​​of different parts of the image are shown in Figures 4-7. Figure 4 shows the changes in the horizontal gray values ​​of the lower and upper edges of the characters in the license plate area (shown by the white lines in the figure). The change of the horizontal gray value of the lower and upper edges of the outside of the license plate area. According to the change of the gray value, the character area of ​​the license plate can be located, as shown in FIG. 6. 2 Segmentation of characters In the license plate area after image segmentation, the internal grayscale of the characters and the bottom of the license plate is relatively uniform, and there is a large difference in grayscale between the characters and the background color, and there is a clear gap between the characters. According to this feature, the number of character pixels in the character area is projected vertically to perform character segmentation. Set a threshold T to distinguish characters from the background color of the license plate, hhi (0≤i
The Spike Light provides a simple outdoor ground lighting solution,which is amazingly flexible and has excellent quality with sleek good looks.Widely used for outdoor, pathway, lawn, outer wall, parterre, garden, villas, hotels, picture lamps, accent lights, porches, pond, deck, etc.The Spike Light are made from many different types of superior materials.You can choose from stainless steel, black, copper or bronze finishes, warm, neutral or cool white light, and different output levels.The light with ground spike is also suitable as wall luminaire for residential and commercial application.Supplied with our plug and cable, simply plug this into an appropriate socket or hard wire into an existing circuit.Moreover,certifications for using in U.S., Canada and all other countries worldwide are available.If you want to know more, please contact us.
Spike Light,Garden Spike Lights,Outdoor Spike Lights,Spike Light Fixture JIANGMEN LEDER LIGHTING CO., LTD , https://www.lederlight.com
1 License plate positioning License plate positioning is the core of the license plate recognition system. It is to find the area where the license plate is located from a complex background image. In order to better locate the license plate, the acquired image needs to be segmented first.
1.1 Image segmentation using improved fuzzy C-means clustering algorithm Fuzzy C-means (FCM) algorithm is a commonly used image segmentation method. It optimizes the objective function by iterative method to achieve image segmentation. The shortcomings of this algorithm are The convergence speed is slow. In order to improve the speed of the algorithm, different improved FCM algorithms have been proposed. In [5], hierarchical clustering is used to divide the image data into a certain number of similarly colored subsets to increase the speed of the FCM algorithm. The improved algorithm improves the speed of clustering by reducing the clustering samples.
In the FCM algorithm, the selection of the initial clustering center and the number of clusters has a certain effect on the speed of the algorithm. A better initial value helps to increase the speed of clustering. The cluster center and the number of clusters are associated with the extreme points of the grayscale histogram of the image. For a more complex image, the grayscale histogram is not a continuous graph. There are many glitches in the histogram, and there are usually many extreme points determined. In order to obtain its extreme point more effectively, we do the following processing on the gray value of the image, superimposing the number of pixels between gray values ​​[h, h + n], where n is the gray interval, This can prevent some extreme points with small pixel values ​​from appearing. Through the processed image gray value col [i] (where 0≤i≤255), the extreme point of the gray histogram is obtained. When col [i-1] 
In equation (1a), when the maximum membership value uRi is greater than the threshold β1, uRi = 1, and the hard C clustering algorithm is changed; when uRi is less than the threshold β2, no correction is made; when β1≥uRj> β 2 At this time, uRj increases to the original 2-uRj times, increasing its membership. In equation (1b), other membership degrees are modified accordingly to satisfy 
The specific operation steps of the improved fuzzy C-means clustering algorithm are as follows:
(1) Perform grayscale processing on the image to obtain the extreme points and the number of grayscale values ​​to initialize the clustering center. The initial clustering center V (0) and the number of clusters C, and specifically select ε> 0 to make the number of iterations k = 0.
(2) Calculate U (K), if ∨j, r, drj (k)> 0, then 
If j, r exists, so that drj (k) = 0, then let urj (k) = 1, and i ≠r, uij (k) = 0.
(3) According to equations (1a) and (1b), modify the membership matrix U (k). 
For the original image shown in Figure 1, the processed gray histogram is shown in Figure 2. The number of extremum points obtained (that is, the number of clusters) is 4, and the cluster center gray value feature quantity is initialized The values ​​are (21, 66, 141, 186). After improving the FCM algorithm, the segmented image is obtained as shown in Figure 3. 
According to the value of hhi, when 
The characters divided according to equation (2) are shown in FIG. 8, and the division effect is more obvious.
3 Conclusion This paper proposes an improved fuzzy C-means clustering algorithm for segmenting license plate images. In the improved algorithm, the image gray histogram is used to initialize the cluster centers and the number of clusters, and the degree of membership in the cluster Corrected accordingly. In the license plate positioning and character segmentation, combining the distribution characteristics of the characters in the license plate, the positioning of the license plate is realized according to the change curve of the horizontal gray value; according to the vertical projection of the number of character pixels in the character area, the character segmentation is realized. The algorithm in this paper is implemented by VC ++ 6.0 programming, and experiments are carried out on multiple images of vehicles with complex background. The experimental results show that the algorithm can obtain the automatic recognition effect of the license plate faster and more accurately.