Image Retrieval Based on Text and Visual Content Using Neural Networks

In the last few years there has been a dramatic increase in the amount of visual data to be searched and retrieved. Typically, images are described by their textual content (TBIR) or by their visual features (CBIR). However, these approaches still present many problems. The hybrid approach was recently introduced, combining both characteristics to improve the benefits of using text and visual content separately. In this work we examine the use of the Self Organizing Maps for content-based image indexing and retrieval. We propose a scoring function which eliminates irrelevant images from the results and we also introduce a SOM variant (ParBSOM) that reduces training and retrieval times. The application of these techniques to the hybrid approach improved computational results.