Based On Improved Color Model And CPSO-AdaBoost Algorithm For Face Detection

With the development of machine intelligence and automation, the status of facerecognition in the field of pattern recognition gradually highlighted with its superiorrecognition performance. Followed by the development performance requirements ofthe recognition system is increasingly high. The ability to adapt to complexbackground environment requirements is getting more demanding. Face detection isthe first part of face recognition system. It refers to the use of correlation algorithmsto mark out its position and size in the input video or image.Commonly used algorithms for face detection, always flawed, such as, trainingtime is too long, detection rate is low and false alarm rate is high. Some algorithmjoining other optimization is able to bring some performance improvements. However,the defects of optimization algorithm will be brought to the detection algorithm. Theuse of AdaBoost algorithms is a fast and effective method. Through systematicunderstanding to the training ideas and detection methods of AdaBoost algorithm,discovery that the performance of face detection system training by AdaBoostalgorithms has further room for improvement. Therefore, this paper is devoted toimprove the training and detection method of AdaBoost algorithm.AdaBoost classifier training algorithm features search time complexity is high,researchers propose an improved algorithm using the best features of PSO-searchmethods, which can reduce training time-consuming, but in the iterative process it iseasy to fall into local optima. This paper proposes the use of chaotic particle swarmoptimization AdaBoost training algorithm——CPSO-AdaBoost. To increase thepopulation diversity by introducing chaos optimization sequence, it expands thesearch scope and helps particle overcome "inert" to get rid of local optimal solution,so that when training the classifier it can quickly find the better performance of theweak classifiers. The results training classifiers face detection in the sample databasecreation show that, CPSO-AdaBoost algorithm has reduced the number of features needed for the training process, shortened the training time, and effectively improvedthe human face detection rate.This paper firstly uses color extraction method for treating skin seized imageregion extraction during detection of human faces. Common color model extractionmethod is capable of removing a portion of the complex background, but it has existshortcomings such as less effective extraction, and less remove background. Thispaper uses skin regions of pictures which is extracted by comprehensive color modelproposed extraction method as a cascade classifier detection window. After deletingthe complex background, image greatly has reduced the search space classifier,obtained the improvement of the detection rate and also played some help onimproving the detection rate.