Dimension reduction of texture features for image retrieval using hybrid associative neural networks

Current multidimensional indexing structures employed in content based image retrieval systems perform poorly when applied to feature data of high dimensionality. To alleviate this problem, one approach is to reduce the number of dimensions of the image data. The authors present a technique of dimensionality reduction using a neural network that combines heteroassociative and autoassociative functions. We show that besides allowing significant reduction in the number of dimensions, combining these two functions can lead to an improvement in retrieval performance