Location And Correction Of Blurry, Smear And Multi Vehicle Plate Images

The vehicle plate recognition technology is the key of intelligent transportation system, and plate position and correction is the main step, which is a challenging task. After the introduction of current development of vehicle plate recognition in condition of complex background, researches on the new methods are included in this paper. All the work of the paper is described as follows:Firstly, an image restoration method based on chaotic particle swarm optimization and regularized parameter by contourlet transform is realized. Contourlet transform is a new multi-scale geometric distortion. Then the regularization model based on contourlet-domain is established. Chaotic particle swarm optimization is adopted to choose the optimal regularized parameter. At last, the extrmal solution of function is found by making use of the fast and exact ability of chaotic particle swarm optimization.The results show that this method has better visual effects and higher peak signal-to-noise ratio(PSNR).Secondly, an image inpainting method of stain vehicle plate based on dual--tree complex wavelet transform is given. It expands primary wavelet transform to plural-domain, which has the good shift invariance, directional selectivity and perfect reconstruction. Then this method confirms the pixel priority and search the optimal matching patch to fill in the inpainting area, which is based on Criminisi idea. At last, the inpainting image is got by wavelet inverse transform. The results show that this method has higher PSNR than primary wavelet transform.Thirdly, a multi-vehicle plate position method based on improved Sobel operator and morphology is given. The image edge is detected by the improved Sobel operator, and then the edge density chart is generated. Afterward, it is binaryzationed and filtered based on non-linear. Then, the image is dilated by morphology method. The external rectangle of the object is found by connected component analyse, and the final vehicle plate is confirmed areas based on the geometric character of it. The experimental results show that the method has higher relevance ratio.Fourthly, an accurate license plate location method based on Tent chaotic particle swarm optimization and texture features is realized. It adapts two different methods to extract the vehicle plate texture features: (1) get the texture feature vector by three one-dimensional filters, and construct the fitness function by leading the texture coherence into the judgment, which could stand for the plate area accurately; (2) get the texture feature vector by seven different vectors and construct the fitness function based on the important characters of plate area. Then the optimal character parameter with tent chaotic particle swarm optimization by combining the texture features of plate area is found. The results show that this method has lower miss ratio and shorter running time.Fifthly, a vehicle plate tilt correction method based on chaotic particle swarm optimization and character rotation projection is proposed. The projection on horizontal and vertical coordinate is used on character region of the rotated vehicle license plate. Then the projection minimum distance is searched by chaotic particle swarm optimization, which is used to get the optimal vertical tilt and horizontal sheer angle. At last, it rotates the image with the anle to get the corrected result. The results show that this method has higher constringent precision and shorter running time.