The Optimize Research Of Face Recognition For Embedded System

As a kind of biometric technology, Face recognition has been widely used in people's life. With the performance of embedded hardware improving, Face recognition technology has a rapid development in the embedded area. This thesis improves Face recognition algorithm in the Special embedded application environment.Generally speaking, Face Recognition includes two processes: Human Face Detection and Face Classification (narrow face recognition). Firstly,in this thesis, a new Real AdaBoost algorithm based on multi-threshold method has been developed. Real AdaBoost algorithm requests to divide the sample space while the traditional division of equidistant can not reflect distribution of positive and negative samples. Through the selection method of multi-optimization threshold value and combined with the strategy of weak classifier threshold selection in discrete AdaBoost algorithm, the thesis has implement the rational division of sample space . We get a good result in MIT-CBCL test set.Secondly, the thesis optimizes face recognition. The global and local features of the human face have covered the integral and detail information separately. And lately, the trend of face recognition is to band them together. Based on this, the thesis puts forward a face recognition model combining global and local features. We extract the features of the whole face picture with PCA and the subfields with PCA+LDA., and, get the double value of sorting weight value and clustering weight value by constructing different training set for the different feather. The identifying process introduces a two-layer classifier. By combining global and local features, this method puts forward the double weight value to make an obvious point of the contribution of each feature when identifying"intra-personal"or"extra- personal"faces, which improved a lot in identification veracity. This method is adapted to little sort recognition in the embedded system.Finally, an embedded touch screen human-machine interface (HMI) is designed using S3C2410 chip based on ARM. This equipment can be used to control multi-models PLC and data collection part. The thesis makes a combination of the grading of operators that in the industrial control to verify the validity of the improved arithmetic.