A New Practicable Vehicle License Plate Location And Recognition System

As a key topic of automatic control domain and intelligent transport domain, automatic vehicle license plate recognition technique has a high value in academic research and practicable application. With the rapid development of automatic control, computer and communication techniques, new methods and technique have been published in resent year. Their applications in engineering have been shown tremendous powers and huge potentials.The automatic license plate recognition system is composed of vehicle license plate location, character segmentation and character recognition. Vehicle license plate location and character recognition are key steps of the automatic license plate recognition system. The vehicle license plate location, character segmentation and character recognition were deeply researched in this thesis. The main contributions of this thesis are as follows:1. The components, structure and principle had been deeply studied for the sake of more practicable vehicle license plate recognition algorithm could be proposed.2. After analyzing the characteristic of the current vehicle license plate location algorithms, a new fast algorithm on multi-features including texture, edge and color features of the vehicle license plate was proposed. The algorithm choosed the mathematical morphology as the tools of mining the texture and eliminating the noise pollution, used particular texture condition and color condition judgment to classify and cluster the sub-region segmented from the image, and finally accurately located the position of vehicle license plate.3. Based on the accurate location of vehicle license plate, the disadvantageous affect of the character segmentation algorithm caused by the dot between the second character and the third character of the vehicle license plate was analyzed according to the rules and the geometrical features of the characters. A new character segmentation algorithm which was based on the Ostu binarization algorithm and a particular amphi-scan method was proposed. The characters of the plate were primarily segmented using projection-histogram and amphi-scan method. And then the character regions were divided and merged. As a result, the characters of the plate were segmented accurately. 4. After the characters of the plate accurate segmentation, the characters of the plate recognition were researched. After the images of the characters normalization, the features of the characters were extracted by coarse grid, and the BP neural network was used to recognize the characters. According to the arrayed feature of the plates, three types of BP network classifying systems, including Chinese characters classifying system, letter classifying system and alphanumeric characters classifying system, were configured. Every character in license plate was recognized by the corresponding classifying system according to its sequence number in the plate. The recognization result of the whole plate was combined by every output of three classifying systems.5. The system software based vehicle license plate recognition algorithm proposed in this thesis was accomplished by Delphi. In the end, the program was encapsulated as a DLL file. In order to correctly utilize the software, the program framework, functions, scheduling sequence of interface functions and explaintation were introduced.