Research On Key Techniques Of Low-Quality Vehicle License Plate Recognition

In this paper, we discussed the design and implementation of vehicle license plate recognition system. Key techniques and related algorithms of low-quality vehicle license plate processing are mainly covered. A lot of work has done on segmentation of color image, auto location of binary image, and character extraction of low-quality license plate images.Following work has been finished in this thesis:1. We studied the license plate location and character recognition systems at home and abroad. Common methods are summarized. What's more, problems of common projects such as low adaptability, low recognition of low-quality license plate are raised. basic conceptions and basic theory are summarized, crucial theory and crucial algorithms are implemented.2. We explored the techniques of image segmentation to color images, based on which an algorithm of license plate location is raised. The character of the algorithm is that we do not need to analyze different images, directly use clustering. At the beginning, to reduce the complexity during clustering, we fully mine correlations of pixels and statistics of whole image, and introduce vector quantization to do color reduction. And we amply expound the conception of dissimilarity increment and its excellent statistic characteristic in clustering. We implement the algorithm in color image segmentation, and use it in license plate location.3. Because of the disadvantages of color image segmentation, we explored the techniques to binary images. We raise an algorithm which synthesizes the advantage of seed filling and template matching. After pretreatment to color images, the algorithm directly turns them to binary ones, and then labels the objects. In the sets of sub-objects, the algorithm searches the seeds which can provide the characters of vehicle license plates, and takes the seeds as minimum templates, grows up and down, according to the rules. In the end, the algorithm takes the regions to match the basic license plate template. The algorithm can locate the vehicle license plates effectively, and it also can locate license plates more than one.4. It is difficult to get characters from low-quality vehicle license plates using common measures. Thus, we raise a new algorithm to get characters based on knowledge-guided techniques. VORONOI diagram owns fine attributes, using which we can reduce thecomplexity of Euclidean distance transformation from o(n~3) to o(n). Based on EDT matrix of the images, we add other characters such as contrast, area. In this way, we can...