Kernel common discriminant-based multimodal image sensor data classification

Multimodal images (for example, optical image, MR, mammography) are widely used in many practical areas, for example, face recognition, image retrieval, and medical assisted diagnosis. In this paper, we proposed a novel image recognition method of kernel common discriminant based image classification. Firstly, we analyze the limitations of the traditional discriminative common vector (DCV) on the nonlinear feature extraction for image owing to the variations in illuminations. In order to overcome this limitation, we extend DCV with kernel trick with the space isomorphic mapping view in the kernel feature space and develop a two-phase algorithm of KPCA + DCV. The experiments are implemented on WDBC, ORL, YALE, MIAS databases to testify the performance of proposed method.