Shadow Removal from Digital Images using Multi-channel Binarization and Shadow Matting

Shadow removal has developed gradually as a preprocessing step for image classification, object detection, information extraction, etc. Existing techniques used by researchers typically make use of segmentation, deep learning to obtain accurate results, which leads to the high cost of processing the image. This paper presents a comprehensive survey study of shadow detection and removal from images. We present our methodology for improving the process of shadow removal which will use machine learning to train the system with standard dataset available (for eg. SBU->Stony Brook University[12]) and then a test data will be entered by the user via an easy to use user application. Training the images will include converting the image into grayscale, YCbCr and CIE L*a*b* colorspace then performing multi-channel binarization which will convert the image to binary on the basis of a threshold value for shadow detection. This shadow detected image will then be filtered to remove noisy false positive regions. This filtered image will be passed to the Canny edge detection algorithm for detecting edges of shadow and then repainting it by shadow matting technique to remove shadows.