Shadow Detection And Removal In A Single Digital Image

This paper introduces the algorithm of shadow segmentation and shadow removal from a single color digital image. Shadow segmentation is the algorithm of determining the contours of the shadows from a digital image. Shadow removal, which is based on the shadow segmentation, is the algorithm of increasing the brightness of the shadow region, in order to have a good visual effect, as well as the lit region. The shadows with fuzzy boundaries are called soft shadows, and the shadows with clear boundaries are called hard shadows. We give the algorithms for the different types of shadows.For the hard shadow images, previous shadow segmentation models are to compare the intrinsic image and the original image to obtain the shadow boundary information. Then they dilate the shadow boundary to make it continuous through the morphological operations. The dilation of shadow edge makes the accuracy of the shadow segmenta-tion reduced. We combine the intrinsic image and modern image segmentation algorithm instead of morphological operations, and give a novel method for shadow segmentation. The soft shadows, which have blurred boundaries, can be divided into two parts, i.e., the penumbra region and the umbra region. We model the problem of soft shadow segmen-tation as a function approximating problem, and then get the boundaries of penumbra regions.The shadow removal algorithms can be operated in the RGB color space and HSI color space respectively. Each color space has its own advantages for the shadow removal. Shadow removal algorithms and shadow segmentation algorithms are highly related. For the hard shadows, we reduce the shadow factor from the original image, after the shadow segmentation completed. If the visual effect at the boundaries is not satisfactory, inpaint-ing technology can be used for improvement. The shadow algorithm for the umbra region is consistent with the method for the hard shadow. The shadow factor in penumbra region is not a constant, so it is more complex to calculate. We model the shadow factor calcula-tion as an energy minimization problem. Experimental results show the effectiveness of our algorithm.