Color and Shape Based Indexing Using Scale-space Representations

Performance of an image or video database is directly related to indexing performance. Proposed indexing methods usually rely on low-level features such as color, texture and shape information extracted from an image or a key-frame selected within a shot, which are very sensitive to small changes in intensity or lightning conditions. In this study, we show that one can extract more robust and rich index vectors based on color and shape through using scale-space representation. Color and shape-based index vectors are derived by utilizing their behavior throughout the scalespace. Developed query system allows users to perform color-only, shape-only and hybrid queries. From the experimental results, it is found that the proposed method give more powerful and rich representations compared to the methods that only utilize colorbased or shape-based indexes, and could index images with similar color or orientation histograms. Moreover the proposed algorithm yields better and consistent results especially for images with similar content.