Incremental Face-Specific Subspace for Online-Learning Face Recognition

A practical face recognition system is expected to have the ability to learn online to adapt to different variations of the imaging conditions in order to achieve better recognition performance, especially when batch training is impossible. This is commonly achieved by updating the face model incrementally for each face. Based on our previous Face-Specific Subspace (FSS) face recognition method, in this paper, an incremental subspace updating method is further applied to FSS in order to make it have the ability to learn online, which is named Incremental Face Specific Subspace (IFSS). Since in the FSS face recognition method, each individual face is represented as one specific face subspace, therefore, the face model can be updated even only single live example face image is available as well as its corresponding class label. Experiments on the Harvard face database show that better recognition performance can be achieved when more example images are incrementally fed into the IFSS system. Furthermore, Experiments also show that IFSS has comparable performance with batch training FSS for the same training sets, but less computational resource is needed and the training examples need not be stored.