Research Of Face Detection Based On Lighting Pre-processing

Face detection is defined as finding the extent and location of a face or many faces in an image or an image sequence. It's an important research aspect in computer vision and artificial intelligence. For colorized face image obtained in complex lighting environment, ideal detecting effect can not be achieved by the existing face detection methods. In the colorized face image, complexion is one of the most notable features, reducing search scope of face and increasing rate of face detection can be achieved by making use of complexion information effectively, but, because of outer lighting, color offset or highlight and shadow may exist in colorized face image collected, and these are obstacle of heightening the rate of face detection based on complexion. So, research of this thesis can be summarized as follows:(1) Based on a detailed survey of the existing face detection methods, the current face detection method based on skin color segmentation and template matching is improved and a method based on lighting pre-processing, skin model and template matching is proposed for detecting human faces in color images. After lighting pre-processing, skin model is established by skin feature;skin detection and threshold segmentation are executed according to skin model;the chosen face area and face template are matched based on analysis of segment area feature;at last, the matched area is earmarked by rectangle. It is the guide idea of research in this thesis.(2) After comparing and analyzing existing face detection methods based on complexion, a skin model used to detect human faces in this thesis, Gaussian distribution model is proposed. Skin model is established after researching deeply and collecting extensively sample. Results of experiment show that most skin area can be separated from background by the combination of this model and lighting pre-processing. It is the base of veracity of face area feature analysis.(3) The Gray World lighting compensation technique and nonlinear transformation of chroma method are introduced to our algorithm, color offset of detecting image is eliminated and problem of color offset caused by the difference of lamp-house's color is solved to a great extent. Nonlinear color transformation is used to eliminate the influence of highlight and shadow so that the highlight and shadow of face image areresolved partly. The mix of these two methods is the base of high rate of skin color based face detection.(4) A face detection process is designed and the face detection code of color image is completely implemented. Our algorithm was tested and analyzed through determination of key parameters, selection of skin-samples and robustness-testing of the algorithm. Experimental results show that the rate of correct detecting frontal and half-profile face image in real environment is 84% and algorithm is insensitive to the lighting and has high robustness to the pose and expression of face images.