Research Of Face Detection Technology Based On Skin Color And Independent Component Analysis

A method based on skin color and independent component analysis is used to detect faces in image with colorful, static and complicated background. The face detection system is divided into three parts, rough detection of face regions, extraction of independent component, and face detection with FastlCA. First, I establish skin color histogram involved chroma and lightness information in UCS color space, and segment likelihood skin color region from colorful image by statistic model of skin color, and then confirm the probable regions containing human faces after eliminating noise and analyzing the figure of the region. Second, I improve the FastICA software package of Helsinki University to extract the independent facial features and detect face, then select the independent components by genetic algorithm and form the space of face image. Finally, I project the probable regions containing human faces to the space of face image and calculate the coefficient of projection, and testify whether the probable face regions are true faces by the nearest neighbor rule and measures the angle between a test vector and a training one.Experiment is carried on in the test face database (164 color images, 203 faces) in Pentium 868MHz, 256MB computer. The hit-rate is 93.1%, the number of false detection is seven, and average time of the whole algorithm is 5.8 second. The results show that, with fewer constrains, the algorithm based on skin color and independent component analysis has a high hit-rate and low false-alarm-rate in face detection on the condition of different lightness conditions, sizes of face, facial expressions, variable orientations, partial occlusion, and complicated background.