Incremental learning algorithm for face recognition using DCT

Abstract

Face Recognition System developed in this research work is based on newly invented Incremental Support Vector Machines for face recognition in which DCT (Discrete Cosine Transform) is used for the purpose to reduce the dimensionality of face space. Low frequency DCT coefficients are used to generate local features. Selected feature vectors are then fed into ISVM to classify the input data as a face ID or not. Incremental Support Vector Machine is used to learn data incrementally from previous stored data and also to avoid large training time and memory consumption for face recognition. In this approach ORL (Olivetti Research Laboratory)[28] face database is used for performing experiments and the results has proved that not only the training time but also the updating time taken by Incremental SVM is very less. Using this technique an accurate face recognition system is developed and tested and the performance found is efficient. The biggest advantage of using the ISVM is that it not only decreases the training time and updating time but also improves the classification accuracy rate to 100%.