Performance Improvement of Vector Quantization by Using Threshold Hung-Yi Chang 1 , Pi-Chung Wang 2 , Rong-Chang Chen 3 , and Shuo-Cheng Hu 4 1 Department of Information Management I-Shou University, Ta-Hsu Hsiang, Kaohsiung County, Taiwan 840, R.O.C. leorean@isu.edu.tw 2 Institute of Computer Science and Information Technology 3 Department of Logistics Engineering and Management National Taichung Institute of Technology, Taichung, Taiwan 404, R.O.C. {abu,rcchens}@ntit.edu.tw 4 Department of Information Management Ming-Hsin University of Science and Technology, Hsinchu, 304 Taiwan, R.O.C. schu@mis.must.edu.tw Abstract. Vector quantization (VQ) is an elementary technique for im- age compression. However, the complexity of searching the nearest code- word in a codebook is time-consuming. In this work, we improve the performance of VQ by adopting the concept of THRESHOLD. Our con- cept utilizes the positional information to represent the geometric re- lation within codewords. With the new concept, the lookup procedure only need to calculate Euclidean distance for codewords which are within the threshold, thus sifts candidate codewords easily. Our scheme is sim- ple and suitable for hardware implementation. Moreover, the scheme is a plug-in which can cooperate with existing schemes to further fasten search speed. The effectiveness of the proposed scheme is further demon- strated through experiments. In the experimental results, the proposed scheme can reduce 64% computation with only an extra storage of 512 bytes. 1 Introduction Currently, images have been widely used in computer communications. The sizes of images are usually huge and need to be compressed efficiently for storage and transmission. Vector quantization (VQ) is an important technique for image compression, and has been proven to be simple and efficient [1]. VQ can be defined as a mapping from k-dimensional Euclidean space into a finite subset C of R k . The finite set C is known as the codebook and C = {c i |i =1, 2,...,N }, where c i is a codeword and N is the codebook size. To compress an image, VQ comprises two functions: an encoder and a de- coder. The VQ encoder first divides the image into N w × N h blocks (or vectors). Let the block size be k (k = w × h), then each block is a k-dimensional vector. VQ selects an appropriate codeword c q =[c q(0) ,c q(1) ,...,c q(k−1) ] for each image K. Aizawa, Y. Nakamura, and S. Satoh (Eds.): PCM 2004, LNCS 3333, pp. 647–654, 2004. c  Springer-Verlag Berlin Heidelberg 2004