Generic Face Detection Algorithm Based On Gray Level, Gradient And Color Characteristics

In this article.we studied general purpose face detection algorithm, put forward a method that used gradation , gradient and skin colour information together. In the face detection method of Viola-Jones.they use haar-like feather to describe the gradation difference rule of face pattern, therefrom learned the rule that can differentiate face / none-face pattern the most with AdaBoost algorithm.forming an cascades classifier for face detection. Hit rate of the mothod is fairly better, and speed is the quickest in the now available method. Enlightened by them.we used haar-like feather to describe gradient difference and the skin color difference rules respectively. We combinated rules of the gradation difference and the gradient and skin colour difference ,design out three kinds of rule sets.they are:G rule set that forms by rules of gradationdifference alone; G~2 rule set that forms by rules of gradation difference andgradient difference together; G~2S rule set that forms by all there kind of rules.Preceding two kinds of rule sets can be used for gray image face detection,but final kind of rule set can use for face detection in color image. Train out three cascade classifier with AdaBoost's algorithm on these three kinds of rulesets,call them G classifier , G~2 classifier and G~2S classifier respectively. By way of testing, under the circumstances that the false alarm rate is identical ,find that the hit rate of G~2 and G~2S classifier exceeds 3% and 5%respectively about than G classifier;The speed of G2 is quick one times almostthan G's ,and speed of G~2S is slightly quick than G~2's.lt is knowable from this,the classifier jointly using three kinds of information mentioned above is better than classifier using gradation information alone.