الفهرس 
INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 80 |  |  |
| | | from | 80 | | |
Abstract
Image clustering is extensively used in many computer vision tasks including Content Based Image Retrieval (CBIR). CBIR is a technique for retrieving images that are most similar in content to a query image as characterized by its visual features from a collection of images. Many techniques have been proposed for image clustering including unsupervised learning techniques that only use unlabeled data to partition images into different clusters and semi-supervised learning techniques that incorporate background information in the form of partial image labels or pair-wise constraints between images to semi-supervise the clustering process. Semi-supervised clustering techniques proved to be successful in overcoming some of the drawbacks of the unsupervised clustering techniques and provide better performance than these techniques. In this thesis, a novel active Affinity Propagation algorithm for pairwise constrained image clustering is proposed. It actively selects the most informative image pairs based on an entropy measure of node uncertainty and then queries human expert for pairwise must-link and cannot-link constraints between these pairs. The constraints are then used as partial background information to supervise the Affinity Propagation based image clustering resulting in a significant performance improvement in the clustering results. Experimental results on different image datasets show that the proposed approach outperforms baseline and state-of-the-art active image clustering approaches in terms of the Jaccard Coefficient achieved with respects to the number of constraints used. This approach continuously achieved higher performance than the other approaches as more constraints are obtained showing its effectiveness in obtaining the most informative pairwise constraints.