| This thesis reports results on the use of shape features for content-based image retrieval. Edges, lines, corners, and straight ribbons are extracted from a query image. These features and their attributes, along with various relations are used to match representations of other images in a database via a graph-matching algorithm. Images can be matched to images or to hand-drawn sketches.; Experimental results on images with man made structures and aerial images are promising. Images can be matched to images of hand drawn sketches. In addition to the content-based retrieval, the matching method is also applicable to object recognition and localization. The algorithm is also suitable for estimation of scene changes in a continuous movie. Presently, the algorithm is developed in MATLAB. The current program must be translated to ‘C/C++’ in order to run much faster and should be tested on a much larger set of images. |
