Research On Image Shadow Elimination Algorithm Based On Color Consistency

With the progress of science and the development of technology, more and moremultimedia information is accessing to people’s lives, and among them, the image informationis a kind of visual information that we always get. Image is an effective carrier of informationand people can get a lot from the image, such as a target like car or people and such like objects.But at the same time, there may also be the interference information in the image, such asshadows.In computer vision, the existence of shadows in the image can reduce the reliability ofmany computer vision algorithms, as well as affecting some computer tasks, including imagesegmentation, object detection, scene analysis, and object tracking etc. So shadow detection andremoval is an important preprocessing work to improve the performance of such visual tasks asmentioned above.As we study and analyze some algorithms about image shadow detection and removal,wefind a problem that although there are many algorithms for image shadow removal, but no onecan removal the shadows completely. We find that the shadow regions whose shadows hadbeen removed cannot be harmony into the whole image. This paper presents an algorithm ofimage shadow removal based on color consistency, which makes an color adjustment to theshadow part after shadow removal and that helps a lot for the region fusion to the wholeimage. In this paper, we do color consistency processing in the border of the shadow region andthe shadow part, we re-coloring on the shadow region and its boundaries in order to achievebetter results. The experimental results demonstrate the effectiveness of the algorithm.First of all, we use an algorithm to do bilateral filtering for the image, and then calculatethe gradient values of this filtered image. Later we’ll do graph-cut by watershed algorithm, andthen use accuracy edge detector to detect the fuzzy shadow edge. I n our paper, delimiter forshadow or non-shadow is denoted as y (y=1represents shadow,-1represents non-shadow), andwe will define adjacent boundary by using a CRF equation. A bad result of watershedsegmentation is that it may generate boundaries in smooth regions of the image. To make upthis, we retain only those with strong edges of the image and we use the Canny operator to dealwith fuzzy shadow edge (with weak boundaries).After shadow detection, this paper uses a simple illumination model to construct the lightenvironment of image. Then determine the coefficient for shadow eliminating to recover thelight information of the shadow region. In order to make the shadow color integrate better in thewhole image, we use the relevant research of color compatibility to processing the image after shadow removal in this paper. Here, we do color consistency processing at the edge of theshadow, re-color the shadow region boundaries and the regions that surrounded by theboundary. This improved the visual effect. In simple terms, we adjust the color of the targetobject (here is the shadow region) to make it more consistency with the whole image.The main content of this paper is divided into six parts. The first part is the researchbackground of the topic and the research status at home and abroad; the second part is theintroduction of the basic knowledge that used in the process of shadow removal, whichincluding the shadow properties, optical knowledge and the color theory; the third part is thepresent shadow detection methods and the method we used in this paper; the fourth part is theintroduction of classical algorithms for shadow elimination and the algorithm we used in thispaper; the fifth part is the experimental results of this paper and the last part is the summarize ofthe advantage and disadvantage of this paper, and the emphasis on the prospects for the futureresearch.