Reducing the dimensions of texture features for image retrieval using multilayer neural networks

This paper presents neural network-based dimension reduction of texture fea tures in content-based image retrieval. In particular, we highlight the use fulness of hetero-associative neural networks to this task, and also propos e a scheme to combine the hetero-associative and auto associative functions . A multichannel Gabor-filtering approach is used to derive 30-dimensional texture features from a see of homogeneous texture images. Multi-layer feed forward neural networks are then trained to reduce the number of feature di mensions. Our results show that the methods lead to a reduction of up to 30 % while keeping or even improving the performance of similarity ranking. Th is has the benefit of alleviating the ill-effects of the high dimensionalit y of features in current image indexing methods and resulting in significan t speeding up retrieval rates. Results using principal component analysis a re also provided for comparison.