An image compression and indexing system using neural networks

A powerful neural network system and its parallel computing implementation is presented in this paper for both image compression and indexing. With image compression, a distortion-equalized fuzzy competitive learning algorithm is developed for direct vector quantization of input images. The output of the neural network is then used to construct histograms for image indexing in which a weighted counting of codewords scheme is proposed to reflect and classify the image contents, Experiments show that substantially improved performance has been achieved with the proposed system for both image compression and indexing which provides promising potentials for further research on real-time indexing of compressed image databases. (C) 1997 Academic Press.