Competitive learning clustering for information retrieval in image databases

Efficient and accurate information retrieval is a key issue in image databases. Since image databases use image features for retrieval, traditional alphanumeric indexing methods are not particularly suitable for content-based retrieval. Therefore, new indexing methods must be designed and implemented specifically for image retrieval. In this paper, we propose to use the competitive learning clustering algorithm to produce an indexing structure for Montage, which is an image database supporting content-based retrieval using color, texture, sketch, and shape for Hong Kong's fashion, textile, and clothing industry. Competitive learning is a stochastic and efficient clustering method which provides good cluster center approximation for image database indexing. Using synthetic data, we demonstrate the Recall and Precision performance of nearest-neighbor feature retrieval based on the indexing structure generated by competitive learning clustering and show that the algorithm works well.