A method for shadow invariant image

The lighting variance brought great trouble for computer vision and has been extensively researched. In this paper, we propose a new method that uses a RGB image to generate a grayscale image that is not sensitive to shadows. We first roughly detect shadows from an image, then we used the detected shadow to generate paired regions of shadows and non-shadows. Finally we used these paired regions to estimate the parameters of the proposed shadow invariant algorithm. Our method does not use any other prior knowledge except just a RGB image. It is simple and effective.