Xiaoqun Zhang, 上海交通大学


Retinex by Higher order Total variation L^1 decomposition


In this paper we propose a reflectance and illumination decomposition model for Retinex based on high order total variation and L^1 decomposition. Based on the observation that illumination varies smoother than reflectance features, we propose a convex variational model which can effectively decompose the gradient field of observed image into salient edges and illumination field by using first and second order total variation regularization. The proposed model can be efficiently solved by a primal–dual splitting method. The tests performed on both gray scale and color images show the strength of the proposed model for applications to Retinex illusions, medical image bias field removal and color correction.

« Back