Shadow removal based on invariant image with Fisher discrimination criterion

Invariant image is widely used to remove shadows in images, however, the proposed methods based on invariant image are too difficult to implement. In this paper, a simple method is proposed in this field. First, the Fisher discrimination criterion is applied to find the invariant direction accurately, and then the corresponding invariant image can be obtained. Second, the linear least squares fitting technique is used to model the linear relationship between the original grayscale image and the invariant image. Then a best-fitting invariant image relative to the original grayscale image can be obtained by using the linear relationship derived before. Note that the best-fitting invariant image has been normalized to the same level with the original grayscale image, so it can use the features of the grayscale image to recover the shadow-free image directly. Finally, the shadow-free image can be recovered by applying the best-fitting invariant image. Experimental results show that this method can remove shadows well in the real scene images.