Research on Double-weight Face Recognition Model

This paper puts forward a face recognition model combining global and local features which is adapted to small sorts face recognition in the embedded system. The mainly contribution of this paper is to bring out the thought of doing the identification using double-weight. Firstly, extract the features in the whole face picture and the subfields; then, construct different training set for each feature to get double-weight value that noted sorting weight value and clustering weight value; the identifying process is a two layers cascade classifier. By combining global and local features, this method is robust to the changes of expression and angle of human face; the bringing out of double-weight also makes an obvious point to the contribution of each feature when identifying "intra-personal" or "extra-personal" faces, which improved a lot in identification veracity. In the experiments, we use the most convenient method called principal component analysis (PCA) to extract features, and the identified veracity has been improved a lot compared to other models based on PCA.