Research Of Restoration Method For Motion Blur Image In Video Surveillance

The restoration technology of motion blurred image, an important research of image processing field, is widely used in the field of video surveillance. Its essence is the best estimate of the blur kernel and sharp images based on blurred image or more prior knowledge and acquisition. The method based on L1/L2 regularized could estimate the blurred kernel accurately and restore image quickly, but there is serious ringing effect in the restored image. To solve this problem, this study presents a revising method based on self-similarity of image. Besides, another improving method of deblur based on the norm Lo is presented to deal with the problem of fuzzy license plates appearing in video surveillance. The main work of this thesis is as follows:(1)The study introduces the background as well as the significance of the image de-blurring, discusses the image degradation model, and concludes the current development of the image de-blurring technology. It also points out the difficulties and key issues of image de-blurring. It elaborates the de-blurring method based on L1/L2 regularization and reveals the shortcoming of the ring effect in the restored image. Then the thesis presents a new method which is based on self-similarity of image to recover fuzzy image.In estimation kernel, the shocking filter is used to strengthen edges of the processing image, which improves the result of estimation kernel. Then, to suppress the ringing noise, it adopts non-local mean filter in blurred image restoration. At the same time, this method is used to de-blur the fuzzy image of pedestrians in video surveillance, thus increasing the identification of pedestrians.(2)The thesis analyses the features of the license plate image, such as histogram of pixel intensities and gradients, and points out the deficiency of Lo regularization based on the intensity and gradient that of image in debluring the license plate image. That is because the pixel values of the license plate image do not focus on zero, and the license plate image is denser than text image. Therefore, the image of license plate is preprocessed in order to increase the sparsity of it and then the method of Lo regularization based on the strength and gradient of image is used to deblur it, by which a clear license plate can be restored.