Content-based retrieval using a multi-objective genetic algorithm

Content-based retrieval from multimedia databases is an important multimedia research area where traditional keyword-based approaches are not adequate. Multimedia data is significantly different from alphanumeric data because multimedia data is generally meaningless to a human and multimedia objects are typically large. Moreover, the traditional keyword-based approaches require an enormous amount of human effort during manual annotation and maintaining the consistency of annotations throughout database evolution. Research on content-based retrieval focus on using low-level features like color and texture for image representation, and a geometric framework of distances in the feature space for similarity. However, systematic retrieval of the best matches in a large multimedia database requires exhaustive and exponential search and does not guarantee worst-case performance. In addition, it has been observed that certain image representation schemes perform better than others under certain query situations, and these schemes should be somehow integrated and adjusted on the fly to facilitate effective and efficient image retrieval. Some work has been done applying simple genetic algorithms for content-based retrieval to provide good, but not necessary optimal solutions. However, these simple genetic algorithms can find only one optimum solution in a single run. This research proposes a new content-based retrieval method based on a multi-objective genetic algorithm (MOGA), which is capable of finding multiple trade-off solutions in one run and providing a natural way for integrating multiple image representation schemes. This research focuses on structural similarity framework that addresses topological, directional and distance relations of image objects.