A comprehensive assessment system to optimize the overlap in DCT-HMM for face recognition

The Hidden Markov Model trained by Discrete Cosine Transform (DCT-HMM) is a very established method for face recognition. However, traditional ways to judge whether the model is a good model is usually one-sided. In Computation time or error rate, researchers usually consider one of the following: (1) to reduce the error rate or (2) to save the computation time. This paper proposes a novel assessment index based on entropy method by considering these two indexes together to evaluate the DCT-HMM system comprehensively. Also, since the block sampling part is important in the process of DCT-HMM, the overlap between consecutive blocks can be optimized by yielding the best assessment index value.