VECTOR QUANTIZATION WITH VARIABLE-PRECISION CLASSIFICATION

We investigate variable-precision classification (VPC) for speeding ve ctor quantization (VQ). VPC evaluates bit-serially, from the most sign ificant bit. When the magnitude of the error due to the unevaluated bi ts is less than the absolute magnitude of the discriminant, we can cla ssify without processing the remaining bits. A proof shows that as the operand precision increases, the average necessary precision becomes asymptotically independent of the operand precision, VPC makes the com plexity of L(2) norm equivalent to L(1) norm. In VQ of real images, on average, the codevector element's precision necessary for classificat ion was under four bits. We implemented binary classification circuitr y using VPC and conventional approaches. The key modules were designed and their performance estimated assuming 1.0-mu gate array technology . The implementations could search binary pruned trees at the televisi on quality video rate. When the overall execution time is important, V PC more than halves the computational complexity.