IRUS: image retrieval using shape

Finding shapes in image databases is a challenging topic in content based retrieval. In this paper the goal is to find database images which contain shapes similar to the query of the user. Unlike most solutions to this problem, the algorithm presented in this paper is meant to cope with changes in rotation, scale, translation, and lossy compression noise. A Java application was built which uses snakes and invariant moments. The GVF snake was used because it has two significant advantages over the traditional snake formulation. First, the GVF snake can fit into concavities, and second, the GVF snake can fit itself to objects using both expansion and contraction of the snake. The objects in the images were segmented with the active contours, and then invariant moments were calculated and compared with a minimum distance classifier. Retrieval quality of the system was measured with respect to original images, rotated images, scaled images, noisy images, and combinations of those distortions.