A method for evaluating the performance of content-based image retrieval systems based on subjectively determined similarity between images