Research Of Algorithm Of Restoration Of Haze Image

Fog and haze are integral and unavoidable features of nature which affect our outdoor visual cognition capabilities. In the hazy weather, the light reflected by objects and environment light is absorbed or scattered because of the aerosol particles. The images obtained have low contrast, image blurring, and details information loss, etc. With the growing increase of demand of outdoor visual system, the restoration problem of haze image gradually became a hot spot in the field of computer vision. But the imaging process of objects is very complicated, so it can not been quantitatively described. What's more, the degradation of haze image is proportional to the depth of object, and recovering scene structure from a single image is an ill-posed problem. Therefore, single image restoration is a challenging and urgent hot issue.This dissertations?employs image degraded model and dark channel prior. Under the basis of analyze the image degradation mechanisms we research the problem of haze image clearness. This dissertation proposed the single image dehazing based the space variation bilateral filter, and to a certain extent it solve the common problem in haze image degradation. In addition, this dissertation?also gives quantitative evaluation method for haze restored image. Mainly there are three aspects:Firstly, a graph-based image segmentation method is applied to segment the hazed image. Thus the object which is not discontinuous in depth of scene and discontinuous in two-dimensional space is segment into different areas. Then we employ the blackbody theory and dark channel prior to obtain the coarse transmission map according to the segment result, and the manipulation can effectively refrain from block effect at the depth-discontinuous edges.Secondly,the existing visibility recovery model is analyzed, and the defect of model is obtained. So the bilateral filter in color space, position space and transmission map space is designed to guarantee the local smooth and contrast enhancement. Therefore, the refined transmission map and good dehaze result are obtained.At last, according to the particularity of haze-degraded image, this dissertation?proposed the representation of image contrast based on the spatial frequency of the image, and the representation of structural similarity based on the edges. Following, on this basis, the image quality assessment algorithm based on the image contrast and structure is proposed. And then we can employ the algorithm to assess, compare and analyze the state-of-the-art haze remove method.