Improvement of Color Quantization Using Self-Organizing Map for Image Segmentation

Abstract

This paper presents the color quantization using a self-organizing map (SOM) and image segmentation. We try to implement perceptual segmentation of color images using only the color quantization and without using the texture features. We propose an improved hierarchical SOM that is a modified version of the Ong's method. The proposed method achieves more stable color quantization results than the conventional method. We also propose to use the pre-processing including the edge preserving smoothing by an anisotropic diffusion that is suitable for reducing small regions and realizing perceptual segmentation.