Fractal based image coding has been shown to work well. The main reason is the ability to capture much signi?cant information while discarding most of the redundancy. Therefore, a similar theoretical apparatus can be used to design a system that extracts information suitable for content based image indexing.
After studying the basics of partitioned iterated function systems as used in image processing, the structure of a fractal based image indexing system is defined by underlining how it evolved and developed over time, going from the image coding-compression stage through a histogram based approach (first and fire, previous algorithms) to a more sophisticated and complex system (fine) that includes Peano-serialized spatial addressing, a linearized image space, a custom clustering strategy, ad-hoc search improving heuristics and specially de?ned distance functions. The resulting system is invariant or robust to a large class of typical variations that appear in natural images including rotations, scaling, and changes in color or illumination. The performance of fine is discussed and compared with other contemporary alternatives using standard and custom-based image databases, mostly of single objects lying against a uniform background. Finally, some possible future developments are designed with the ultimate goal of being able to deal with more complex pictorial scenes.