A hybrid approach for improving Content Based Image Retrieval systems

This paper presents a hybrid image retrieval system which integrates Neural Network and Genetic Algorithm together. Proposed method reduces the semantic imbalance between the machine description and the human semantics of an image by using low-level feature descriptors- HSV color histograms, color moments, and wavelet transform, which matches human perception. These descriptors when used to train the neural network are successful in learning the semantic class of images to a great extent. Genetic algorithm is employed to add on more relevant images to the retrieval results by optimizing the threshold value used for displaying similar images. Implementation results show improved precision and recall for the proposed method.