Use of Self-Organizing Maps for texture feature selection in content-based image retrieval
Chen Guo, Campbell Wilson · 2008
The “Semantic Gap” observed in content-based image retrieval (CBIR) has become a highly active research topic in last twenty years, and it is widely accepted that domain specification is one of the most effective methods of addressing this problem. However, along with the challenge of making a CBIR system specific to a particular domain comes the challenge of making those features object dependent. Independent Component Analysis (ICA) is a powerful tool for detecting underlying texture features in images. However, features detected in this way often contain groups of features which are essentially shifted or rotated versions of each other. Thus, a method of dimensionality reduction that takes this self-similarity into account is required. In this paper, we proposed a Self-Organizing Map (SOM) based clustering method to reduce the dimensionality of feature space. This method comprises two phases: clustering as well as representative selection. The result of the implementation confirms this method offers effective CBIR dimensionality reduction when using the ICA method of texture feature extraction.