High-quality image deconvolution is required for many image processing applications.Our work concentrates on portraying a new image deconvolution method based on Retinex prior knowledge.We build a new image degradation mode,on the one hand,the data fitting term considers a spatial mask that keeps only pixels that do not depend on the unknown boundaries,on the other hand,the regularization term considers the image prior information and a new cross-channel prior(a prior knowledge of Retinex)information.Then the extended alternating directions method of multipliers(ADMM)is used to solve the image deconvolution convex optimization problem.In experiments,compared with several state-of-the-art deconvolution algorithms,including BM3D,HYP and YUV,our method can not only obtain a color aberration correction effectively to better cope with degraded images and get high-quality restoration results,but also achieve stronger applicability.