Facial shape and albedo reconstruction based on a trained prototype

Facial shape and albedo can be estimated under unknown lighting conditions according to a facial prototype, which is generated by using Photometric Stereo. However, the prototype is seriously affected by shadows, ambient light, and noises. A solution to this problem is to train the prototype based on a facial database. In our algorithm, the facial shape and the albedo of a novel human face under unknown lighting conditions are reconstructed based on the trained prototype and singular value decomposition. Extensive experiments have proven that using our proposed trained prototype is more robust than other methods when a face is under shadow.