Quaternion-based image shadow removal

Visual effects such as shadows and highlights appear in photographs due to variations in lighting conditions. Although these effects add more meaning to the images, they pose various problems to specific computer vision algorithms. Hence, the removal of shadows and highlights is often considered as a prerequisite to such algorithms. This paper presents an interactive technique for shadow elimination from images. Our method requires user input in the form of rough strokes on the shadow region and its corresponding non-shadow region in the image. We further use quaternion rotation in the YCbCr color space to derive an image that is invariant to shadows. The actual colors of the image are finally recovered by color transfer from the original shadow image. The proposed method takes less time to generate the shadow-free image and does not necessitate the detection of shadows prior to its removal. Also, unlike the existing shadow-removal techniques, our method generates invariant image with minor texture loss. Experimental findings are reported to demonstrate the performance of our shadow-removal technique.