Abstract: To avoid the halo artifacts and color distortion by the traditional center/surround-based Retinex enhancement methods,this paper proposed a Retinex method based on improved Mean Shift filtering．Firstly,low contrast image is decomposed into two parts:luminance and chrominance by principal component analysis (PCA)．Improved Mean Shift filtering is introduced to get the enhancement of luminance adaptively．In addition,color image inpainting is performed independently according to Meylan method.Finally,image compensation is completed based on the global analysis．The experimental results show that the proposed method can effectively overcome the halo artifacts while maintaining color consistency when dealing with low contrast images,and speed up the process comparing with the similar adaptive methods.

Key words: Retinex,Image enhancement,Mean shift,Principal component analysis