Face Detection Algorithm Based On Improved YCrCb Color Space

Face detection is very important step of face recognition system. It is a relatively complex cross-disciplinary subject, involving pattern recognition, digital image processing, algorithm design, computer graphics and other related disciplines, the result of face detection provides direct data for face recognition to process, so the quality of face detection determines the recognition rate and the recognition speed of face recognition, and the accuracy rate of face detection is the key to improve the recognition rate and the recognition speed of face recognition.The author has read a large number of academic papers and reports which publish in recent years on face detection, found the accuracy rate of face detection is influenced by the illumination strength, and bring out a face detection algorithm which base on improved YCrCb color space, finish a face detection system and get a good detection result. This article has some good study result which list as follows:First of all, improve traditional convert algorithm From RGB color space to YCrCb color space.Traditional convert function from RGB color space to YCrCb color space is a linear function, By some experiment author find that the different of the face color'Cr value and Cb value between image under high illumination condition and image under normal illumination condition, and the value of Cr and Cb tend to zero with the increase of illumination strength. If the range of illumination is big the value of Cr and Cb are decentralized, so skin color model is useless to detect face in image with variable illumination condition. Author analyze the tradition convert function and the geometry meaning of YCrCb color space and RGB color space, and find a non-linear convert function convert from RGB to YCrCb. And prove that the value of Cr and Cb which get by improve function is more centralizedSecondly bring out face detection method combine projection and template matching.This thesis use projection and template matching to filter candidate face, resize candidate face to fixed dimension, and carve up the candidate face into several same sizes partial in horizontal direction and vertical direction. Compute the number of each partial create projection to find the distributing character of pixels in the candidate face. Compare it with stand face template's projection to find if they have same character. In addition, use the color distributing character of face to template matching, if candidate match template more exact then decide it is face.At last, finish the face detection system which based on improved YCrCb.Experiment proves that this method can reduce the influence of the illumination strength for color information based face detection, and improve the accuracy of face detection and it's efficiency is better than traditional method.