Research On License Plate Recognition Algorithm In Complex Natural Environments

License plate recognition is an important part of the intelligent transportation system.It is one of the important research subjects of computer vision,image processing and pattern recognition in the field of intelligent transportation.But the license plate image collected in the actual environment that is easily affected by many adverse factors such as,illumination change,scale change,target interference,so license plate recognition is still a very challenging task in the complex nature.License plate recognition technology mainly solves the three problems of license plate location,segmentation and recognition.In this paper,the three parts are studied respectively,and the corresponding algorithms are proposed.In this paper,Based on target regions a license plate location algorithm is proposed,and a progressive refinement location strategy is adopted.The algorithm is applicable to complex natural environment such as illumination change,scale change and target disturbance.This paper introduces the Selective Search algorithm for object region extraction of the input image,according to the characteristics of license plate region candidates were screened,and the candidate regions are distinguished by a pre-trained support vector machine,retaining plate area.The non maximum suppression(NMS)is used in license plate region to eliminate the coincident region.Finally,the location of the license plate is accurately located.A character segmentation algorithm based on connected region is proposed in this paper.Firstly,the input plate is pre-processed and slant corrected,and the character region is obtained by means of connected region labeling and mathematical morphology processing.After that,the traditional method of character normalization is improved,and the problem of character deformation caused by normalization is solved effectively.In this paper,a license plate character recognition algorithm based on convolutional neural network is proposed.Two convolutional networks NET1 and NET2 are designed.NET1 is used to recognize Chinese characters,and NET2 is used to identify letters and numbers.This paper introduces rectifier as the activation function of neurons,and uses mini-batch stochastic gradient descent method to train the network,which can accelerate the convergence of the objective function.The convolution neural network can automatically extract the image features from the input character images and classify them,so as to obtain the recognition results.Throughout the process,you do not need to manually select image features or do local processing of images.Experiments show that these algorithms can effectively locate the license plate in the complex natural environment,segmentation of characters and recognition characters.These algorithms is compared with the same type of algorithm,and all of them are improved significantly.