Face Detection Using Quantized Skin Color Regions Merging and Wavelet Packet Analysis

C. Garcia, G. Tziritas, University of Crete, 1999

Detecting and recognizing human faces automat-ically in digital images strongly enhance content-based videoindexing systems. In this paper, a novel scheme for human facesdetection in color images under nonconstrained scene conditions,such as the presence of a complex background and uncontrolledillumination, is presented. Color clustering and filtering usingapproximations of the YCbCr and HSV skin color subspaces areapplied on the original image, providing quantized skin colorregions. A merging stage is then iteratively performed on the setof homogeneous skin color regions in the color quantized image,in order to provide a set of potential face areas. Constraintsrelated to shape and size of faces are applied, and face intensitytexture is analyzed by performing a wavelet packet decompositionon each face area candidate in order to detect human faces. Thewavelet coefficients of the band filtered images characterize theface texture and a set of simple statistical deviations is extractedin order to form compact and meaningful feature vectors. Then,an efficient and reliable probabilistic metric derived from theBhattacharrya distance is used in order to classify the extractedfeature vectors into face or nonface areas, using some prototypeface area vectors, acquired in a previous training stage.