ABSTRACT
This chapter presents a new technique for object recognition using singular value decomposition and principal component analysis (SVD-PCA). The implementation of InfoMAX is publicly available MATLAB code written by Bell and Sejnowski, and is used to generate the results. Although PCA is intrinsically very close, SVD differs in important aspects that can affect the performance of the classifier used in the recognition system. The result is compared with other standard transforms, like PCA and independent component analysis (ICA). The potential of SVD-PCA on a database of 7200 images of 100 different color objects is illustrated. Overall, it is observed that SVD-PCA performs significantly better than conventional and subsequent eigenvalue decompositions. Experimental results in appearance-based object recognition confirm that SVD-PCA offers better recognition rates over ICA and PCA. The excellent recognition rates achieved in all of the experiments performed indicates that the proposed method is well-suited for 3D object recognition in applications like surveillance, robot vision, biometrics, and security tasks, etc.
