An Empirical Analysis of Digital Image Color Quantization using Partition Based Clustering Techniques

In today’s technical world, dependency on digital devices as well as the generation of digital data has increased by many times. Out of the total data produced through various digital devices, image data occupies a large section. Devices can create images in different formats such as monochrome, gray-scale, and color. Color images again can have different color models out of which the red green blue (RGB) model is one of the most widely used models. The total number of colors that can be present in an RGB color image can be as high as 224 = 16777216. All devices may not be able to efficiently process these high numbers of colors and most importantly, all these colors may not be of much significance from the perspective of human vision or various types of applications. In this paper, we have used partition-based clustering algorithms such as K-Means++, K-Medoids, Fuzzy-C Means, and Self-Organizing Maps to reduce the number of colors in RGB images from 16777216 to as small as 512 (i.e from 224 to 29) colors. According to the experimental results, K-Means++ clustering gives better performance in comparison to the other three clustering methods for color quantization.