Shadow Removal using Multi-Channel Binarization, Color-Line Clustering and Illumination Estimation
Saiqa Khan, Meera Narvekar, Taniya Fansupkar, Umama Maghrabi · 2021
Shadows are a typical aspect of pictures and once left undiscovered will slow down scene understanding and visual processing. Existing methods adopted by researchers based on segmentation and deep learning lead to high processing time. In this paper, we have proposed a technique for shadow detection and removal which involves performing clustering using color-line identified by performing offset correction, and then estimation for illumination is carried out. After that, an automatic shadow detection method is used. Then, the shadow scale is estimated by the ratio of shadow-free areas. At last, illumination optimization is used to improve the shadow scale. Detection of shadows involves color space conversion and binarization using a threshold for different channels of an image. The output image will be further processed to remove noisy regions. The method used for shadow detection is automatic which overcomes the manual work of the user to select the shadow region and based on this selection, segmentation was performed which consumed more time. Comparatively, our approach for shadow detection is expeditious and thus reduces manual work.