A Face Recognition Method Of Support Vector Machine Learning By MapReduce Model

Face recognition as a typical biometric authentication, has become a veryimportant research field of pattern recognition, and has broad application prospects. Inrecent years, with the rapid development of mobile Internet, the face recognitionapplication generated new demand. But because of the large amount of computation,high demands on the memory capacity, battery life and other hardware in the mobileenvironment, the traditional face recognition method is difficult to cope with the largeamount of data processing in the mobile Internet environment.Face recognition based on statistical features is the most common method, in orderto research recognition rate of face recognition using classifier based on the statisticallearning theory, this paper has researched face recognition method based on supportvector machine and neural networks, Bagging algorithm point out that recognition bytraining multiple support vector machines and then train a support vector machine fromthem could get a better recognition rate. MapReduce model is firstly proposed byGoogle, the model can support large-scale data parallel computing, which is easy toimplement and has good scalability. At the same time, using the MapReduce modelcould implement the Bagging algorithm. Based on the above reasons, this paperproposes a face recognition method whisc using MapReduce model to train supportvector machines for face recognition under Hadoop platform. In order to solve themulti-class classification problem like face recognition, we proposes a train methodusing MapReduce model and “one against all” or “one against one” multi-classclassification method. In this method, multiple support vector machines are generatedfrom input sample slices in the Map stage, and then the support vectors meet theconditions will be given to Reducer for next process to get the support vector machineused to recognition. The experiment on three sets of samples with different size showthat the train speed of OAO method and OAA method has improved by usingMapReduce model, the greater the number of samples, the superior performance of thetwo methods on the recognition rate and training speed, in addition, OAO method ismore stable than OAA method on recognition rate.The innovation of this paper is using the MapReduce model to train support vectormachine for face recognition, and proposing train strategy of multi-class support vectormachine using MapReduce model. The experiment show that using MapReduce model to train support vector machine for face recognition has reduced the burden on hardwareand improved efficiency of computation under the premise of ensuring the recognitionrate. The method has practical significance for face recognition in mobile enviroments.