Algorithms for fast full nearest neighbour search on unstructured codebooks: a comparative study [articol]

Abstract

This paper presents several fast nearest neighbor search algorithms for vector quantization on unstructured codebooks of arbitrary size and vector dimension that uses linear projections and variance of a vector. Several new inequalities based on orthonormal Tchebichef moments and projections on the first vectors of the DCT and PCA transformations of an image block are introduced to reject those codewords that are impossible to be the nearest codevector and cannot be rejected by inequalities based on Hadamard Transform, sum and variance, thereby saving a great deal of computational time, while introducing no extra distortion compared to the conventional full search algorithm.