Content Based Image Retrieval for Unsegmented Images

We present a new method for image retrieval by shape similarity able to deal with real images with not uniform background and possible touching/occluding objects. First of all we perform a sketch-driven segmentation of the scene by means of a Deformation Tolerant version of the Generalized Hough Transform (DTGHT). Using the DTGHT we select in the image some candidate segments to be matched with the user sketch. The candidate segments are then matched with the sketch checking the consistency of the corresponding shapes. Finally, background segments are used in order to inhibit the recognition process when they cannot be perceptually separated from the object.