An unsupervised face detection based on skin color and geometric information

A robust algorithm is proposed for face detection in a complex scene in which parts of some faces are overlapped one another, some faces are posed differently, and image contains many different patterns. The algorithm starts from extraction of skin pixels and then segments facial regions overlapping into several separate ones. The divided region is determined whether it contains a face or not by geometric relationship between the extracted lip and eye components. Lip segmentation is proposed based on wavelet multi-scale edge detection across discrete Hartley transform. Eyes are determined with the help of cues of the extracted lip components. Comparative study with some existing face detection algorithms has indicated the superior performance of the developed algorithm from experimental results