Retinex theory is a kind of technology for image enhancement, and its basic principle is to decompose the original image into the product of illumination and reflectance. Traditional Retinex theory generally assumes that illumination is smooth, while in actual scenes, because of effects of different reflection characteristics of targets and shadings, illumination is piecewise smooth, resulting in an edge blurred reflection image and generating halos. For the above problems, this paper proposes an improved image processing framework for Retinex theory. First, the illumination is estimated with piecewise smooth operator to eliminate the halo phenomenon. Second, illumination correction is introduced for adjusting the contrast of the image. Third, reflection correction is introduced for detail enhancement or denoising. In the end, this paper verifies the improved framework with image quality evaluating indicators for the validity and correctness. The experimental results show that the improved framework is effective.