Auto Location And Recognition Of Car License Plate

The car plate recognition system is an important component of Intelligent Transportation System. Research on cars license plate recognition system(LPRS) involves techniques such as digital image processing,computer seeing,pattern recognition and artificial intelligence etc. LPRS plays an important role in intelligent traffic control system, and it can be applied in vehicle management in situations of all levels and all kinds.LPRS consists of three modules in general,they are license plate location,character segmentation and character recognition. Algorithms of all modules related to LPRS are deeply studied and analyzed in this paper. And,we finally settle down a series of algorithms used in LPRS based on our research.Car license plate locatiion is the key process in car license plate recognition system. The efficiency of the location will directly affect the precision of recognition. In this paper we proposed a new method for car license plate location by combining morphology and multiple features. Firstly, a color image is turned into gray image and its edge is detected by using the Sobel operator; then, some morphology operators are used to delete noise and form some connecting areas; finally, the plates are located by length to width ratios and the number of white pixels of every connecting areas. The experimental results show that the method used in this paper can detect car license plates effectively and fast.In character segmentation, we gave a method which is based on projection and pre-known knowledge of the character size and the space between characters in plate.In character recognition, an improved BP neural network is used to carry out the recognition of letters and numbers in the license plate. Ideal and noised characters are individually used to train the network so that the recognition system has better capability. BP algorithm is one of the most widely used algorithms in neural network. But, through analyzing the main idea of it, some weak points of it are found, such as easy to fall into local minimum value and slow in convergence. An improved method,Gradient descent with momentum and adaptive learning rate back-propagation,are introduced to restrain these weak points. The experimental result shows that the method used in this paper can enhance the learning speed and improve the recognition precision.