Halfway through the semantic gap: Prosemantic features for image retrieval

We present here, an image description approach based on prosemantic features. These features are obtained through a two-level feature extraction process. A first level of features, related to image structure and color distribution, is extracted from the images, and used as input to a bank of classifiers, each one trained to recognize a given category. Packing together the scores, the features that we call prosemantic are obtained, and used to index images in an image retrieval system where searches are performed using relevance feedback. Prosemantic features have been evaluated on a public domain dataset, and compared against two different sets of features. Our experiments show that the use of prosemantic features allows for a more successful and quick retrieval with respect to the other features considered.