A novel atmospheric turbulence-degraded image restoration algorithm based on support vector regression
Chunsheng Liu, Ming Li · Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE · 2006
A novel method based on support vector regression is presented for atmospheric turbulence-degraded image restoration. Firstly, an operation with a sliding window is employed to the images to analyze the correlation between pixels of clear image and 8 neighbors of corresponding pixels of degraded image. After feature selection, we get training samples. Secondly, an appropriate kernel function is employed to map the training samples into a higher space. Through linear learning machine in kernel feature space, we get non-linear function. Then the relationship between clear images and degraded images is constructed via regression analysis of the training samples by a support vector machine. Thus the model for turbulence-degraded image restoration is constructed here. Finally, the degraded images to be tested are restored by this model. The experimental results show that the proposed method has lower NMSE and higher PSNR and runs faster than classical image restoration methods such as Wiener Filter, Iterative Blind Deconvolution and etc.