In this study, a novel face recognition framework based on the grammatical face models has been proposed to tackle partial occlusion problem. The grammatical face model represents a face by five isolated 'subface', forehead, eyes, nose, mouth and chin models in cooperation with 'occlusion' models. With the creations of 'subface' and 'occlusion' models, the authors then define a facial grammar to manipulate 'subface' and 'occlusion' models for constructing various composite face models structurally. Furthermore, the authors also introduce a universal 'occlusion' model, which could handle general occlusions to improve the robustness and flexibility of grammatical face models. The proposed face recognition system could overcome two problems. One is to resolve the problem of face recognition with partial occlusions; the other is to overcome a challenge of training face models from occluded face images only. Experimental results carried out on AR facial database reveal that the proposed approach outperforms the state-of-the-art methods.
All Science Journal Classification (ASJC) codes
- Signal Processing
- Computer Vision and Pattern Recognition