Research On Shadow Removal In Remote Sensing Images Based On Classification Compensation

With the rapid development of science and technology, satellite remote sensing ushered in an unprecedented new stage. Remote sensing technology have been applied in more and more fields due to cannot be matched by other technology. However, In the remote sensing image, because there are a lot of tall buildings and trees will block the sun’s rays, lead to form the shadow on the ground. The existence of the shadow not only has a beneficial effect but also has a bad influence on the application of remote sensing image. We can through the shadow’s size, shape and direction to estimate the size of the obstructions’shape and the direction of the light source, etc. But, the reduction or even loss of information in the shadow regions causes problems some applocations in the field of computer, such as object recognition, image segmentation, image registration and so on. Therefore, recovery the color and textures information of shadow region to reduce or even eliminate the adverse effect which will make the remote sensing image data could be fully and properly used. Recovering the information of shadow region contains shadow detection and shadow removal.In this paper, we will research on the shadow removal.This paper describes the principle and the experimental simulation analysis of existing shadow removal algorithm, in view of the shadow removal method based on nonlocal regularization still exist color information can’t recovery enough good, a shadow removal method based on nonlocal regularization of bright dark compensation was proposed. Firstly, the method divide the shadow and the corresponding similar unshaded region into bright area and dark area, through executing classify compensation respectively to get a predicted image which can maintain color information better. And then, We analyze the difference of shadow scale in the R,G,B three channel dimension of different ground in shadow region, therefore put forward the R,G,B three channel dimensions should be allocated weight value respectively of the shadow scale nonlocal regularization to reduce the color distortion while enhance the detail information. The experimental results show that the improved method can better restore color information of shadow region.In view of the method based on nonlocal regularization of bright dark compensation have low efficiency and color distortion shortcomings, a method combining classification compensation and high-pass filtering was proposed. Firstly, the method through outline to find each nonshadow class which corresponding to shadow class, shadow class with the corresponding nonshadow class one to one mapping compensate to weaken the influence between class and class. Then through gaussian high-pass filter on the shadow image and improve the brightness of filter results, which get a more detail infomation image. Finally combining classification compensation results with high pass filtering results to enhance the texture and color information of shadow region. The experimental results show that the proposed method not noly can maintain texture and color information well but also improve the efficiency of the algorithm.