A two-layer framework for appearance based recognition using spatial and discriminant influences

Appearance of objects lie in high-dimensional spaces. Feature selection improves not only the efficiency of object recognition but also the recognition accuracy. In this paper, we propose a two-layer learning framework of feature selection using spatial and discriminant influences. The first layer selects a number of feature points of highest integrated influences by integrating spatial and discriminant influences, and the second layer refines the selection in terms of the discriminancy of these feature points measured by orientation histograms of their local appearances. The proposed framework can be categorized as a global appearance based recognition approach. Unlike popular projection methods, such as PCA, LDA, the proposed framework can present visual interpretability of selected features, which is desirable in bioinformatics and medicine informatics. We present two case studies: (i) embryo stage recognition and (ii) face recognition. Our case studies show the effectiveness of the proposed framework.