| Summary: | 碩士 === 國立東華大學 === 資訊工程學系 === 90 === The goal of this paper is to improve the recognition of multi-orientation faces. In previous approaches for face recognition, although there exist many different methods for feature extractions and classifications, most of them belong to one-stage recognition. This kind of approach usually causes interference between samples with large variations. Hence, this paper proposes a two-stage method for multiple orientations.
The main concept of two-stage recognition is to recognize faces through coarse and fine classifications. The coarse classification, which is usually termed as clustering, is separate face patterns into different clusters according to their orientations. The fine classification is to accurately identify the face patterns as the target individuals. The benefit of this approach is the avoidance of interferences between multi-orientation faces.
Our experimental results on two databases show that the two-stage recognition indeed improves the recognition rate over the one-stage methods. In the future, we will expand the method to multi-stage recognition for face images with multiple expressions and under varying light conditions based on the same concepts.
|
|---|
