GDL

Face Recognition Using Invariance with a Single Training Sample

For the limits of memories and computing performance of current intelligent terminals, it is necessary to develop some strategies which can keep the balance of the accuracy and running time for face recognition. The purpose of the work in this paper is to find the invariant features of facial images and represent each subject with only one training sample for face recognition. We propose a two-layer hierarchical model, called invariance model, and its corresponding algorithms to keep the balance of accuracy, storage and running time. Especially, we take advantages of wavelet transformations and invariant moments to obtain the key features as well as reduce dimensions of feature data based on the cognitive rules of human brains. Furthermore, we improve usual pooling methods, e.g. max pooling and average pooling, and propose the weighted pooling method to reduce dimensions with no effect on accuracy, which let storage requirement and recognition time greatly decrease. The simulation results show that the proposed method does better than some typical and nearly-proposed algorithms in balancing the accuracy and running time.

Beri Komentar ?#(0) | Bookmark