Numeral Recognition Based On Image Processing Technology

Within the last two decades, with the rapid development of computer vision technology, the character recognition technology has been applied to more and more areas. For example, license plate recognition, billing identification input, house number identification, books scanning recognition and so on. In the face of large amounts of data that need to identify input, manually becomes inefficient and has high error rate. Although at present the market has already appeared in many character recognition systems, many of them can only recognize single case of characters one or image information relatively simple and standard characters. The reality is often more complex, different sizes of characters, noisy background of characters, tilt angle of characters have effect on recognition efficiency and accuracy.This thesis research design of this situation,a really viable, able to solve some of the complex circumstances of the license plate number recognition system. The system of repeated experiments, the comparison of neural network algorithm and template matching algorithm, in both efficiency and accuracy rate of identifying cases selected template matching algorithm more practical, and one by one in the program implementation. The system includes:the image pre-processing, license plate location, character recognition most of these three. Contrast in the feature extraction stage through several algorithms used in this paper efficient and high recognition rate by pixel scanning feature extraction method, which makes following the BP network training can converge quickly. But for real-time require a higher degree of accuracy demanding and complex environment of network training seems less efficient. Thus, when confronted with a series of license plate number recognition practical problems, the use of template matching will yield better recognition effect.The experiment proved that the system design of the paper plate number for identification, both in Chinese characters, letters and numbers correctly on the rate of 90% or more, the recognition speed in 2 seconds, and the complex environment with high adaptability is of practical significance.