Shadow Detection and Removal from Real Images

Abstract

Shadow detection and removal is used in various image processing applications like video surveillance, scene interpretation and object recognition. Ignoring the existence of shadows in images may cause serious problems like object merging, object lose, misinterpretation and alternation of object shape in various visual processing applications like segmentation, scene analysis and tracking. Many algorithms have been proposed in the literature, that deals with shadow detection and removal from images as well as videos. A comparative and empirical evaluation of the existing approaches in video has already been reported, but we lack a similar one in case of still images. This paper presents a comprehensive survey of existing shadow detection and removal algorithms reported in the case of still images. Evaluation metrics involved in shadow detection and removal techniques are discussed and the inefficiency of conventional metrics such as; per pixel accuracy, Precision, Recall, FScore etc in detection phase are also explored. Quantitative and qualitative evaluation of selected methods are also discussed. To the best of our knowledge this is the first article that exclusively discusses shadow detection and removal methodologies from real images.