Study On The Algorithms For Single Natural Image Restoration

Image restoration is one of the most popular topics and problems in the area of image processing. Due to the influence of weather and man-made conditions such as shaking of cameras during image acquisition, images are easily been degraded by noise or haze. Image restoration is proposed to solve the above problems by building the corresponding degenerative models. The degraded images can be restored based on the reverse process of each degenerative model. Image restoration is widely used in the field of intelligent transportation systems, such as intelligent transportation, medical imaging, space exploration, remote sensing, military affairs and criminal identification.Image deblurring and image haze removal have both achieved big breakthrough thanks to the advanced study of imaging technology in these years. However, there exist few researchers who focus on the images which have both blur and haze included in. Thus, it is the consideration of this paper. This essay starts by an analysis of the popular algorithms of image dehazing and image deblurring. Then a new algorithm is proposed based on the statistical characteristics of the intersted images to restore the images which have blur and haze at the same time. First, a haze removal algorithm is developed based on the method of dark channel prior. Promising and faster performance can be achieved comparing with the older methods since it utilizes the guided image filter and refines the dark channel image to improve the transmission map. Moreover, by statistical analysis of the obtained dehazed image, we found that the dehazed image still follow heavy-tailed distribution. Then, with the characteristic of the images, a method to restore the degraded images by both blur and haze is proposed. In this paper, the dehazed images with blur are handled based on the variational Bayesian, and the RL deconvolution algorithm is improved to induce the ringing artifacts around sharp intensity transitions availably. It can be proved from the experimental results that the proposed approach can achieve better visual results.