An efficient mettiod of constructing L/sub 1/-type norm feature to estimate euclidean distance for fast vector quantization

In order to speed up the search process of vector quantization (VQ). i t is inost important to avoid computing k-dimcnsionat Euclidcan distance as many as possible. In order to f-ind a best-matched codeword (winner) in the codebook for a cerlain input vector, i t is a general way lo roughly estimate other than exactly compute Euclidean tlistaiice iintnediately for the piirpose of rejecting a candidate codeword. The lower dimensional features of a vector siich as sitin or the mean (L, norm) and LZ nonn are widely used for this purpose. Obviously, how Io construct a suitable feature is a core probtem for estimating Euclidean disiance. In this paper, an efficient niethod of constructing L,-type nonn I'eature i s proposed by introdwing a reference vcctor. 111 addition. the criterion on how to select an optimal referencc vector i s 3ko given. Experimental rcsults confirined the elrecliveness or the proposed incthod.