EPSRC Grant M33051-01

We developed a perceptual approach to generating features for use in indexing and retrieving images from an image database. The concept is that salient regions that immediately attract the eye are large colour regions that usually dominate an image. Features derived from these will allow search for images that are similar perceptually. We compute colour features and Gabor colour texture features on regions obtained from a multiscale representation of the image, generated by a multiband smoothing algorithm based on human psychophysical measurements of colour appearance. The combined feature vector is then used for indexing all salient regions of an image. When retrieving images, those images are selected that contain more similar regions to the image the query came from by using a multi-feedback retrieval and ranking mechanism. Matches are found using the L_2 metric. The results demonstrate that the proposed method performs very well.

A comprehensive Java-based user interface to this system is under developement. Information on this will be posted here soon.

• Set I: This is a query on the face of the subject.
• Set II: This is a query on the night sky but the multifeedback system also picks up on the "white" regions of the tower.

Please see the papers below for further discussion of the results.

• M. Mirmehdi and R. Perissamy. Perceptual Image Indexing and Retrieval. Journal of Visual Communication and Image Representation,volume 13 (4): 460--475, December 2002.
• M. Mirmehdi and R. Perissamy. CBIR with Perceptual Region Features. In: Proceedings of the 12th British Machine Vision Conference, T Cootes and C Taylor, editors, pages 511--520. BMVA Press, September 2001. Acrobat:777456 bytes.