Face recognition using shading-based curvature attributes

In this paper, we apply shape-from-shading to face images to extract fields of surface normals and make estimates of surface curvature attributes. The attributes studied include minimum and maximum curvature, mean and Gaussian curvature, and, curvedness and shape-index. The curvature attributes are encoded as histograms and are used to perform recognition using a number of distance and similarity measures including the Euclidean distance, the Shannon entropy, the Renyi entropy and the Tsallis entropy. We compare the results obtained using the different curvature attributes and the different entropy measures. Using precision-recall curves, we find that the best performance is delivered when the Shannon entropy is applied to histograms of shape index and curvedness.