Research Of Algorithm Of Restoration By Haze Image

In bad weather, the image of outdoor scene is degraded by absorbed orscattered because of the aerosol particles. The current visual system in the designcannot do with the disturbance of the adverse weather, generally only suitable forfine weather conditions. So the adverse weather restricted the promotion of thevisual system. Fog is a common weather phenomenon, and image degradationcaused by it became a hot spot in the field of computer vision. Therefore, thisdissertation is research of algorithm of restoration by a single haze image.This dissertation employs retinex theory and image degraded model. Weresearch the problem of haze image clearness in homogeneous and heterogeneousfog. In homogeneous fog circumstance, this dissertation proposed retinex dehazingalgorithm based guided image filter. In heterogeneous fog circumstance, theparticle swarm optimization algorithms are proposed for the single image dehazing,and to a certain extent it solve heterogeneous fog problem. In addition, thisdissertation also gives quantitative evaluation method for haze restored image.Mainly there are three aspects:Firstly, this dissertation proposed retinex dehazing algorithm based guidedimage filter to solve fog degraded problem. A commonly encountered artifact withretinex algorithms is the existence of hallows, noise and interative solution. Tosolve these problems, a method by two guided image filter is applied to realizeretinex enhancement and a multi-resolution framework combines guided imagefiltering and wavelet thresholding is presented.Secondly, the particle swarm dehaze algorithms is proposed, because thecurrent dehaze algorithm result is bad for heterogeneous fog. First of all, derivationheterogeneous fog degraded model. Then, the particle swarm algorithm is used toestimate to parameters in recovery model. The last, good dehaze result are obtainedby guided image filter.Thirdly, according to different dehaze algorithm, this dissertation proposed anobjective evaluation criteria. Although image quality assessment has beenextensively studied, few methods can be applied in dehaze assessment. So theimage quality assessment algorithm based on the image contrast and structuralsimilarity is proposed, and we can employ the algorithm to assess and analyze thestate-of-the-art haze remove method.