Enhanced Incremental Bi-Directional Principal Component Analysis with Forgetting Factors

Feature extraction plays an important role in face recognition system as it can reduce dimensions and reserve the most significant features which need to be classified and recognized. Principal Component Analysis (PCA) has been one of the popular techniques that used in pattern recognition related research areas. Researches have been also carried out to improve the performance of this technique, mainly based on tensor type and incremental type. Incremental Bi-Directional Principal Component Analysis (IBDPCA) is one of the latest improved versions of PCA which combined the merits from tensor and incremental type. However, IBDPCA lacks of the moderations between the latest and previous data when updating the means. This can leads to difficulty in evaluating the data accurately due to larger size of previous data, and also more memory waste. This paper proposed a technique which overcomes the limitations by adopting the IBDPCA with forgetting factors, in order to downweight the previous overloaded data with relevant factors. To evaluate the proposed technique, two experiments were carried out to compare it with IBDPCA on two different databases: FERET and CMU PIE. The experiment results indicate the better performance in recognition rate by using the proposed