The Degaded Image Retoration Technology Research Based On Non-local Means

The research of how to deal with the degraded image and make it clearer has important realistic significance, because it can be applied to many fields, such as astronomy sensing, military spy, vision traffic surveillance and control system, police and safety system, and diagnostic system etc. Image restoration is a hot topic in the domains of image processing, computer vision, and pattern recognition. It consists mainly of denoising, deblurring, restoration and super-resolution, while the single image defog and the single image super-resolution reconstruction are researched in this paper.Firstly, the principles of the image degradation and reconstruction are analyzed. Two degraded image degradation models such as foggy degradation model and low-resolution image degradation model are investigated. Then the Non-Local means algorithm is introduced.Secondly, the restoration algorithm of foggy degraded images based on Non-Local means filter in condition that the prior information of scene depth is unknown is proposed to solve the problems that the defog algorithm can easily lead to the halo effect of the edge and make the contour and landscape features blurred. First of all, the average treatment is carried out in advance and the sky brightness is estimated with Non-Local means filter. Secondly, the atmospheric veil is estimated based on Non-Local means filtering algorithm according to the fact that the edge of the atmosphere is similar to the high-frequency data of the atmospheric veil of foggy image. And the atmospheric veil is estimated in order to avoid the hard-solved scene depth map. Finally, the smoothing to prevent enlarger of contrast and tone mapping are implemented. Experimental results show that the proposed algorithm can obtain more accurate atmospheric veil, so feature contours and features of the restored image are not only clearer, but also can restrain the halo effect of the edge.Thirdly, the single image super-resolution algorithm based on Non-Local means filter and sparse representation is proposed in order to solve the problem that most super-resolution algorithms are not satisfying in the situation of low SNR circumstances, and most of them are reconstructed based on gray image. First, two over-completed isomorphism dictionaries for the low-resolution and high-resolution image patches are trained based on image patch pair. Secondly, super-resolution reconstruction based on sparse representation and Non-Local means filter is executed in the brightness domain. Lastly, the UV chroma super-resolution reconstruction based on Non-Local means is put forward. The experimental results show that proposed algorithm can not only effectively inhibit edge halo disturbance and artificial artifact, but also naturally robust to noise.