Fast encoding method for vector quantization based on subvector technique with a modified data structure

The encoding process of vector quantization (VQ) is a time bottleneck to its practical applications. In order to speed up the process of VQ encoding, it is possible to estimate Euclidean distance first with just a lighter computation to try to reject a candidate codeword. In order to make an estimation for Euclidean distance, appropriate features of a vector become necessary. By using the famous statistical features of the sum and the variance for a k-dimensional vector and furthermore for its two corresponding (k/2)-dimensional subvectors, it is easy to estimate Euclidean distance so as to reject most of unlikely codewords for a certain input vector as proposed in [2]-[5]. Because it is very heavy to online compute the variance of a k-dimensional vector, a new feature, which is based on the variances of two subvectors, is constructed in this paper for estimating Euclidean distance. Meanwhile, a modified more memory-efficient data structure is proposed for storing all features of a vector to reduce extra memory requirement compared to the latest previous work [5]. Experimental results confirmed that the proposed method in this paper is more search efficient.