Full-Searching-Equivalent Vector Quantization Method Using Orthogonal Projection

The encoding process of vector quantization (VQ) is computational complex and time consuming. Compared with traditional Euclidean distance computation, some geometric properties can generate approximations with simpler computation and filter out impossible codevectors to reduce computation time. In this paper, we introduce a new approximation adopting orthogonal projection on a hyperplane for speed-up VQ encoding. Experimental results show that our proposed scheme requires only 2.7~23.2% of actual Euclidean distance calculation in the full searching. Having been proved, our proposed scheme requires only 30.1~40.3% of computation time that other recently proposed schemes need.