Feature extraction using supervised spectral analysis
Ruicong Zhi, Qiuqi Ruan · 2008
This paper proposes a feature extraction algorithm, called supervised spectral analysis (SSA) which is motivated by spectral clustering. The algorithm is interesting from a number of perspectives: (a) utilize the class information of the data points to construct the affinity matrix, which can enhance the discriminant power of the features; (b) solve the small-sample-size problem which is often confronted in the practical application; (c) effectively discover the nonlinear structure hidden in the data. We analysis the properties of the SSA and apply it to facial expression recognition. Experiments on JAFFE and Cohn-Kanade databases show the effectiveness of the SSA algorithm.