License Plate Recognition Algorithm Based On Wavelet Packet Analysis And BP Neural Networks

The license plate automatic recognition technique is one of the very important technique in modern intelligent transportation system, which is the hot spot of research in the recent years. The intelligent transportation system plays an important role in vehicle tracking, highway automatic charging, parking automatic charging, traffic flow statistics and so on. As the license plate is the only identity of each vehicle, the license plate automatic recognition technique stands at a key role in the modern intelligent transportation, which has fatal research significance. Researchers at home and abroad has done further work on the license plate automatic recognition technique, proposed many algorithms and proposals, and some products have put into use. but they have not achieved the requirements as people as expected, therefore, how to improve the correctness and real-time capacity of the license plate location and recognition algorithm, which have much work to be done.Based on previous research, this paper use wavelet packet transform in the license plate recognition, the main research work include as follow aspects:(1) license plate location; (2) character segmentation; (3) character recognition.In the license plate location aspect, firstly, images carry out gray level transformation, smoothing processing. Based on the method of traditional image enhancement, this paper proposes a method of enhancement algorithm used wavelet packet transform. Combined with edge detection, morphologic processing and projection method, this paper proposes a method of the license plate location algorithm used texture feature. According to the experiment, the algorithm achieves the goal and the rate of the license plate is over 92.9%.In the character segmentation aspect, based on multi-resolution method of wavelet packet transform, this paper carries out mages to noise-suppressed processing, combines with vertical projection and prior knowledge to segment the character. The algorithm is much better than the traditional methods based on projection.In the character recognition aspect, this paper analyses two common method of character recognition:template matching and neural network. This paper proposes that wavelet packet extracts characteristic energy vector as neural network input vector, uses momentum BP neural network of adaptive learning rate to recognize character, which can improve convergence and learning rate of BP neural network.After the experimental verification, the integrated plan which this paper proposes that is effective and feasible, the license plate recognition technique based on wavelet packet analysis and BP neural network has the unique superiority in aspects of speed, recognition rate and so on, and the system has a broad prospect...