Abstract
Human face recognition plays a significant role in security applications for access control and real time video surveillance systems, and robotics. Popular approaches for face recognition, such as principal components analysis (PCA), rely on static datasets where training is carried in a batch-mode on a pre-available image set. Real world applications require that the training set be dynamic of evolving nature where within the framework of continuous learning new training images are continuously added to the original set; this would trigger a costly frequent re-computation of the eigen space representation via repeating an entire batch-based training that includes the new images. Incremental PCA methods allow adding new images and updating the PCA representation, and offer the advantage of dispensing with the recently added images after model update. In this paper, various incremental PCA (IPCA) training and relearning strategies are proposed and applied to the candid covariance-free incremental principal component algorithm. The effect of the number of increments and size of the eigen vectors on the correct rate of recognition are studied. The results suggest that batch PCA is inferior to the four considered IPCAI-4, and that all IPCAs are practically equivalent with IPCA3 yielding slightly better results than the other IPCAs.