An improved kernelized discriminative canonical correlation analysis and its application to gait recognition

Based on the canonical correlation analysis (CCA) and its extended algorithms, an improved kernelized discriminative canonical correlation analysis (KDCCA) was proposed in this paper. Compared with the existing KDCCA, there were two improvements. Firstly, when the kernel method was added, by improving the optimization objective function, the correlation between the final canonical correlation characteristics of the non-corresponding elements were reduced and improved classification results. Secondly, a more general class relationship matrix without sorting the samples was used for adding the class information. Finally, the proposed method was applied to gait recognition to solve the multi-view and different states problem. Experimental results show that the proposed method performs satisfactory recognition results.