SHADOW DETECTION IN A SINGLE IMAGE

A digital image is divided into patches of pixels, or “superpixels”, where each of the pixels in the patch has approximately the same color value and each of the patches is of approximately the same size. Subsequently, eigenvalues are generated for each patch based on the color values of the individual pixels in the patch, as expressed in a multidimensional color-space. A ratio between the first largest eigenvalue and the second largest eigenvalue for each patch is determined and is then, subsequently, normalized, either based on the intensity of the corresponding patch or normalized to a range between zero and one. The resulting values are compared to a threshold to identify those regions of the digital image that are deemed to be in shadow, thereby generating a shadow mask for the digital image, or are weighted to generate a shadow probability mask.