Research On Shadow Processing In Aerial Remote-sensing Images

With high spatial resolution and large information capacity, aerial remote-sensing images have been widely applied in fields of urban economy and social development such as geomatics industry, urban information construction, services and tourism and so on. With the continuous improvement of remote sensing technology, image resolution has grown by hundredfold times. In the meantime, however, the unconspicuous shadow becomes particularly noticeable. Shadow weakens or eliminates the optical and physical information of the shaded area, which brings great difficulties in subsequent image processing including target recognition and ground cover classification. Consequently, research on shadow removal technique has become one of the hottest and the most difficult problems of image processing in massive remote sensing images produced by aerial remote sensing technique. International and domestic academics have proposed many methods for shadow processing in aerial remote sensing images, which implement functions of shadow detection and compensation effectively. On this basis, however, how to increase the accuracy and speed of shadow detection and compensation, and how to expand their applications are one of the branches of research on shadow processing.Analysis of the advantages and disadvantages of shadow detection algorithm based on 3c component when applied to detect shadow in aerial remote sensing images is made, thus the improvements are proposed as follows. First, nonlinear enhancement of 3c image. Optional parameters are added to log function to expand dynamic range, and thresholding method is used to further improve the contrast. Second, image smoothing processing. Weighting formula is defined according to the distance between other pixels and center pixel, and thresholding method is used to revise the image blurring effect of edge detail. Third, edge detection of shadow. The Gaussian distribution and sampling method are used to estimate threshold of shadow edge, which offer the automatic selection of the threshold for the improved Sobel edge detection method. The experiment shows that the proposed method increases the accuracy of shadow detection, reduces the probability of that non-shadowed pixels are wrongly detected as shadow, and is applicable to detect shadow in aerial remote sensing images under various scenarios.Analysis of the advantages and disadvantages of existing shadow compensation algorithm based on color constancy when applied to remove shadow in aerial remote sensing images is made. Aiming at the inaccuracy when calculate gain corrections and neglection of penumbra computation when removing shadow, this paper proposes regional shadow compensation method based on color constancy. First, compensation of umbra. At first, homogenous area is determined by the size of shadowed area and is marked by method of “layer-by-layer traversal”. Then, gain corrections are achieved by color constancy calculation for shadowed area and non-shadowed area respectively. And at last, color of umbra area is restored by applying color consistency formula for each pixel in umbra. Second, compensation of penumbra. At first, extension operate is executed on the result image after shadow edge detection to obtain penumbra area. Then, illumination compensation formula is designed based on piecewise polynomial formula. And at last, color of penumbra area is restored by applying the designed illumination compensation formula for each pixel in penumbra.Because of the specificity of aerial images including diversity and complexity of ground covers, shadow detection and compensation methods of aerial remote sensing images have been the difficulty for image processing. In shadow detection, this paper analyzes common features of shadowed area instead of unique features of each shaded ground cover. Therefore, analyzing the related features of ground covers specifically to detect shadow more effectively is another research direction in future. In shadow compensation, this paper utilizes the color constancy calculation result of homogenous area as non-shadowed area and add penumbra compensation step, however, it is worthy of further research on how to increase the compensation accuracy on that basis.