Image Color Reduction Based on Self-Organizing Maps and Growing Self-Organizing Neural Networks

Color is one of the most important properties for object detection. Color Reduction of Image (CRI) is an important factor for segmentation, compression, presentation and transmission of images. The main purpose of CRI is to cut off the image storage spaces and computation time. Kohonen's Self-Organizing Maps (KSOM) can generate mappings from high-dimensional signal spaces to lower dimensional topological structures. The main characteristics of KSOM are formation of topology preserving feature maps and approximation of input probability distribution. Growing Self-Organizing Neural Network (GSONN) has got more and more attentions in the past decade, to overcome some limitations of KSOM. An effective approach to solve CRI problem is to consider it as a clustering problem and solve it by using some adaptive clustering methods, such as KSOM and GSONN. This paper first gives an introduction to KSOM and neural gas network. Then, we discuss a typical GSONN, growing neural gas. After that, a performance comparison of KSOM and GSONN for CRI is given. It is ended with some conclusions.