Many content-based image retrieval systems retrieve images from a database based on their similarities. But image similarity is both subjective and task-dependent. So it is necessary to learn user's similarity-measure to improve the performance of retrieval. User's similarity-measure is dependent on user's sensitivity to visual features which the retrieval system uses. In this thesis, we propose an algorithm to analyze and apply user's sensitivity to visual features.
We use user's relevance feedback to analyze the user's sensitivity to visual features. User selects the images which, user thinks, are similar to the query from the result of retrieval. Using the distances between query and the selected images in the feature-space, we analyze user's sensitivity to visual features for that query. We make training data composed of feature vector of query and sensitivity vector. We use neural network for system to learn the user's sensitivity to visual features for any query.
We develop content-based image retrieval system. Using this system, user can retrieve general landscape images. This system uses HSI color feature. In this system, we use our algorithm to analyze and apply user's sensitivity to visual features.