Face Detection Based On Skin Segmentation And Improved Adaboost Algorithm

Face detection is the purpose of the research in this paper that to construct a face detection system possess high detection rate and low false detection rate, system integration algorithm combine skin color segmentation with high detection rate and Adaboost algorithm with low false detection rate.Firstly,preliminary locating the faces from the image to be detected with the skin color segmentation,the greatest degree detecting the faces that are possible,then verifying the faces form preliminary position with improved Adaboost algorithm,confirm the position of real face.The specific contents are as follows:Firstly,skin color segmentation is used to complete the preliminary located faces.Establish skin model of BP neural network in the specific color space,the classified image of skin color pixels and non skin color pixels is achieved.Because BP neural network has the disadvantages of easy to fall into local extremum and be sensitive to the parameters,consequently the improved artificial fish swarm algorithm(IAFSA) has been used to optimized the connection weights and threshold of BP neural network.According to IAFSA,an improved algorithm which based on parallel behavior of artificial fish. The convergence speed and accuracy will be raised through the implementation of artificial fish swarming and following behavior happen at the same time.Comparing the experiment of IAFSA-BP and the traditional BP,show that IAFSA-BP has better convergence speed and effect.Secondly, the Adaboost algorithm is improved on the too long training time and excessive distribution of samples.The double threshold is used to construct weak classifiers instead of a single structure to reduce the number of training samples, thus the training time is reduced.The weight update rule is redefined by setting threshold in each round of training,comparing each round of sample weights and thresholds,the sample weights allocation is improved.The improved Adaboost algorithm is used to train classifier for a large number of face feature,so the classifier has good ability to detect face with lower false detection rate.Adaboost classifier is used to verify the face part again for image after color segmentation, face detection accuracy is achieved.Finally,a fusion algorithm based on skin color segmentation and improved Adaboost algorithm would be proposed in the paper.Construct the experiment of skin color segmentation,improved Adaboost algorithm and the fusion algorithm.The fusion algorithm which is presented in this paper shows that it can keep high detection rate of human face,meanwhile the false detection rate reduced.