Mammogram Classification with Local Phase Quantization Features

The potential of local phase quantization for feature extraction from mammogram classification is exploited in this paper. The resulting features are next projected onto lower dimensional space via Fisher linear discriminant analysis. Mammogram split between normal and abnormal (tumor) classes is finally carried out with the help of nonlinear support vector machines. The experiments indicated that the proposed approach was able to provide a satisfactory mean recognition accuracy of 87.78%, outperforming Gabor based feature extraction approach.