Multi-profile Face Detection In Complex Background

Paper studies the detection technologies for multi-profile people face in complex background, including color balance, skin detection, color separation, multi-profile people face location and verification. Adopting theory analysis, experiment to prove etc., compares the algorithms in this text and other, draw the conclusion. The research results are as follows:(1) Color balanced algorithm of gray patch , remedies the deficiency of white patch algorithm invalid in analyzing low gray level picture and the defect of gray world algorithm which can only use for picture with abundant of color variation.(2) RGB-SS skin detection models, as for the function, is combination of color space and complexion model. But setting-up on the basis of being worth picture element for the three-dimensional space coordinate, it is fast to run; Free of luminance, the accuracy is high. There are 50 pictures with no color skew and different illumination carrying on experiment, and examining rate of skin detection with RGB-SS model is 99.4%, examining rate is only 10.5% by mistake.(3) The method of color separation based on edge characteristic , the method on the basis of normal distribution and three color histograms, the application of which makes people face detection procedure change from many people face mode or face picture mode of complicated background to pure single face picture mode.(4) The method of main fact yuans analysis on the basis of weighting of gray level , applies face deflected to correct, basically adapts to these face states such as the low gray level , wearing the glasses or deep deflection slightly.(5) Improved algorithm of eyes location based symmetry transform. Expanding picture region two times, this algorithm is not merely stood for the symmetrical intensity of the selected area on the edge, and the impact on symmetrical intensity of edge expanded only stays on the picture expanding , thus can measure some eyes lying in the edge , and the result is accurate. To experiments of 60 pictures located for the first time, the correct rate of eyes location is 98.3%.(6) Based PCA and SVM, recognize the eyes in order to judge if the selected is one face.