<title>Color vector quantization by competitive learning</title>

Abstract

ABSTRACT In this paper, color vector quantization is performed by a competitive learning based clustering algorithmwith some modifications that eliminate the false colors that may appear on the resulting image. The preliminary operations that must be applied to the input image pixels before the algorithm can be applied are also stated. Moreover, it is demonstrated that with this scheme, faster convergence and less computations are possible usingonly a small fraction of all the pixels, but at the same time producing satisfactory results. Finally the results arecompared to those ofthe K-Means clustering algorithm.Keywords: color vector quantization, clustering, competitive learning, K-Means algorithm 1. INTRODUCTION The need for vector quantization of color images arises when trying to use display or printing devicesefficiently, that can work with only a limited number of colors. A similar problem occurs with the images in thetextile industry, when an original pattern designed by an artist is digitized by a color scanner to have 16 Millioncolors. Before this pattern can be used in production, it has to be quantized to the original number of colors becausethe production line is limited to only a small number of colors.The problem of vector quantization of color images can be expressed in the following way. We consider thepixel as a point in the 3D world, the color components, the R, G, B values ofthis pixel as its x, y, z coordinates. Weare trying to group all of such points in the 3D world into clusters such that within a cluster, the points must be asclose to each other and the distance between different clusters must be as large as possible. After the clustering isdone, the pixels within each cluster are represented by the centroid ofthat cluster.The organization of this paper is as follows: First, the competitive learning algorithm is described and theway it is adapted to the problem is given. Second, the K-Means algorithm is discussed and finally the results aregiven together with the demonstrations.