Multiple-clue face detection algorithm using edge-based feature vectors

A multiple-clue image perception has been developed aiming at direct implementation in a VLSI hardware accelerator, and was applied to the problem of face detection. In the algorithm, feature vectors are generated utilizing distribution of edges in an input image, thus achieving dimensionality reduction for efficient processing. In addition, multiple clues in the edge distribution are utilized to enhance the accuracy of face detection. For this purpose, several feature vector generation schemes have been developed, which are all compatible to the direct hardware implementation. In software simulation, over 91% of human faces have been detected correctly with only 20 face templates, and the number of false positives has been reduced drastically by the multiple-clue scheme.