Rui, Y., Huang, T., and Chang, S. Image retrieval: current techniques, promising directions and open issues. Journal of Visual Communication and Image Representation 10, 4 (April 1999), 39–62. [PDF]
This article summarizes many years of research in the field of image information retrieval. It describes open challenges and the state of the art in the field.
One of the main difficulty results from the rich content in the images and the subjectivity in human perception. This, according to the authors creates a mismatch between the metadata annotations, produced with different techniques and the retrieval efficacy and satisfaction perceived by the user.
To the extent of improving image information retrieval systems, “humans have to be in the loop”. The authors cite a good deal of work that has been conducted to this specific extent (e.g., QBIC interactive region segmentation, the interactive FourEyes, the dynamic feature vector recomputation of WebSEEK, the MARS and PicHunter relevance feedback, and so forth.
The ultimate end user of an image retrieval system is human: therefore the study of human perception of image content from a psychophysical level is crucial.
Additionally, the autors refer to Rogowitz et al. (1998) who conducted a series of experiments analyzing human psychophysical perception of image content. According to their results, even though visual features do niot capture the whole semantic meaning of the images, they do correlate a lot with the semantics. These results encourage the development of metrics to achieve semantically meaningful retrievals.