Improved AdaBoost Face Detection Algorithm Achieved Under S3C2440Platform

Face detection and face recognition is a hot topic in the field of computer vision andartificial intelligence research. Researchers use the face features to identity verification incomparison with other human biological characteristics, such as fingerprint, iris, palmprints, voice, etc., the advantages of the face feature are simple, accurate and user friendly.Face recognition technology make the face recognition safety and convenient. When thistechnology is mature, it will greatly change people’s existing way of life.Embedded system is widely used with its advantages of small size, low powerconsumption, cost advantages. Face detection and recognition algorithms realize inembedded systems, will bring greater competitiveness of its higher cost and otheradvantages. As the bottom embedded systems and hardware platforms has been relativelymature, the chip can realize all kinds of functions, and its speed become faster and faster, ithas stronger performance, and the price of chip is getting lower and lower. Therefore, theembedded system with face detection and face recognition has great practical significanceand application prospects.On the basis of many references, this paper mainly to do the following work:(1)This paper proposes a new weight update rule to improve the problem of thedegradation in the AdaBoost algorithm, and makes some sample weight value notexcessively increase in the case of repeated classification error, so the method avoid theoccurrence of the degradation phenomena.(2)In the face samples using for training people face classifier with AdaBoostalgorithm the number of face samples less than the number of non-face samples, this willresult this problem of imbalance in training data. For this problem, this paper improves theformulas of sample weights normalized, thus the result ensure that in each round oftraining, the total proportion of face sample weights are not too small.(3)The selection process of optimal weak classifier is to use an exhaustive search todetermine the classification thresholds, so that the characteristic values only need evaluate once in the training process, without the need to repeat the calculation, this can furtherreduce the training time.(4)This paper verifies the improved AdaBoost algorithm in the Windows. Theexperiments show that the improve algorithm is better than traditional AdaBoost algorithm,and a new and improved algorithms also runs under Linux, and verifies the effectiveness ofthe new algorithm once again.(5) This paper achieves the goal to improve face detection procedures based on theS3C2440embedded platform, and verify the feasibility of the system.