Computer Vision and Interface Design

In this work, we suggest better ways to present visual information (image databases) for browsing and retrieval. Thumbnails obtained from an image set give a good overview of its contents. Instead of simply downsampling images to obtain thumbnails, we first find salient regions (saliency map) using local statistical features of the image. We crop and downsample the images based on these saliency maps, and obtain better thumbnails. The suggested methods of finding salient regions are faster than existing methods while giving comparable results. Secondly, we have developed a Content Based Image Retrieval (CBIR) system to provide empirical evidence (by user study) that similarity based grouped and hierarchical placement of images is better than random placement. Using an effective shape based similarity measure we conclude that visual search is very useful in image retrieval systems. We conducted a field test to check the robustness of the system in varying photography conditions