A Novel Unsupervised Optimal Discriminant Plane
Suqun Cao, Shitong Wang, Quanyin Zhu · 2011
Optimal discriminant plane is an important feature extraction method in machine learning, pattern recognition and image processing, etc. Based on this, Zhao et al. presented a hybrid optimal discriminant plane method by integrating uncorrelated and orthogonal discriminant vectors together. As the same to optimal discriminant plane, it is a supervised feature extraction method. This paper extends Zhao's optimal discriminant plane to the unsupervised pattern. With the orthogonal constraint and the conjugated orthogonal constraint of the fuzzy total-class scatter matrix, two vectors which maximize the fuzzy Fisher criterion are obtained. These two vectors make up a novel unsupervised optimal discriminant plane. Experimental results on UCI datasets show its efficiency.