The study of feature-level fusion algorithm for multimodal recognition

An improved feature-level fusion algorithm based on kernel canonical correlation analysis is presented and applied to multimodal recognition based on fusion of ear and profile face in this paper. The fusion of ear and face biometrics could fully utilize their connection relationship of physiological location, and possess the advantage of recognizing people without their cooperation. First, only the profile-view face images including ear part were captured for recognition. Then the kernel trick was introduced to canonical correlation analysis, and the improved feature-level fusion algorithm based on KCCA with kernel function optimization is established. With this method, a kind of nonlinear associated feature was proposed for classification and recognition. The results of experiments show that the algorithm is efficient for feature-level fusion, and the ear and profile face based multimodal recognition performs better than ear or profile face unimodal biometric recognition.