Blurred Image Restoration Algorithm Based On Learning

Because of many external influences, image inevitably degrades in the process during forming, transmitting or recording, and always performs in blur, distortion and noise. So image restoration becomes one of the most important and fundamental research fields.Aero-optical effect means that when the high-speed aircraft flies in the atmosphere, the turbulent flow affects the optical imaging system of detection and guidance in heat, radiation and image transmission, so that the object image offsets, dithers or blurs. In order to describe the blurring degree of the aero-optical image intuitively, we develop the software of Image Degradation Tool to calculate the offset, dithering and blurring parameters.Support vector machine is a machine learning method based on Structural Risk Minimization. Based on SVM, we develop the software of Image Restoration Tool. We get the sampling data by sampling the blurring image and the clear image respectively, and then carry out cross validation, consistency testing, and build the image restoration model by training the data at last. We use the trained model to restore the single image or image sequences. Experiments show that it is better than restoration based on winner filter.Image focus is an important evaluation criterion of image restoration. The dual tree complex wavelet transform proposed by Kingsbury, which performs well than discrete wavelet transform, can describe detail information in six directions. Based on the characteristic of multi-directional and analytical, we define a ratio of high-pass band and low-pass band norms as a measure of image focus. Compared with several conventional image focus functions, such as gray variance, Laplace of energy and etc, it performs well either in stability or accuracy.