Genetic Algorithms (GAs) are applied to image browsing system that interactively searches target images, and new GA (QGA) is proposed for this application.
This research project introduces an interactive image browsing system. This system is based on the combination of similarity-based image retrieval technology and simulated breeding methodology, where simulated breeding methodology refers to the generic framework of human interface for the interactive optimization of subjective problems.
To begin with, a user is asked to determine the first example image to start image browsing. Then the system retrieves similar images to the example image and displays images retrieved in higher orders by similarity-based image retrieval. Next the user is requested to input a significance point to each displayed image. This significance point, in turn, is used as the fitness of genes. Then retrieval parameters associated with similarity-based image retrieval are interactively optimized using "queue-based genetic algorithm" (QGA), which this paper proposes as a fit model for simulated breeding.
As stated above, the model and the scenario of image browsing is summarized in this paper. Furthermore, image representation model called "hierarchical model of image content elements" is used for extracting and indexing image contents. Combining the image browsing model and image representation model, Motif-based window interface is constructed for satellite cloud image database of 1027 images. Results of image browsing history demonstrate steady improvement in terms of the similarity retrieval order of the target image by means of both the sequential change of example images and optimization of retrieval parameters by the proposed genetic algorithm.
For your information, the following reference may be one of the definitive reference in this field.
H. Takagi, Interactive Evolutionary Computation: Fusion of the Capacities of EC Optimization and Human Evaluation, Proceedings of the IEEE, Vol. 89, No. 9, pp. 1275-1296, 2001.
I published several papers in Japanese on interactive evolutionary computation applied to content-based image retrieval (CBIR), I regret that I do not have any English papers on this topic. This does not mean that I will not publish English papers for ever on this topic. At this moment, however, my research is a little far away from here, and I probably will not write new English papers without making significant advances in this field. Nevertheless, some people ask me for an English reference on my work, so, just for your convenience, I put here an unpublished manuscript which was submitted to GECCO-99 conference without accepted. Hence this is just a draft, but if you happen to refer to this document, the following is the citation information (Note that addresses in the manuscript has completely changed since then, but I decided to keep the original information).