Real-time Shadow Detection and Removal by Illumination Drop Point Analysis

The existence of shadows in natural scenes cause challenges in computer vision applications such as object detection. In this paper, a proposed approach is introduced for shadow detection in single images. Unlike other approaches, our algorithm uses the variation in the RGB components in order to locate the drop in intensity and analyze it. The input image would be subjected to noise reduction using a Gaussian filter then vertical scanning is applied where the pixels at every column of the image are collected, grouped using a threshold to obtain smooth knee points through the variation of the intensity levels and analyzed. Shadow removal is done and then horizontal scanning is carried out following the same procedure. The analysis and optimization process was done on 300 images and the final testing was done on 4000 images from the SBU shadow images dataset. The results have shown the success of our algorithm to detect high accuracy of shadows than other approaches.