Integrating Hexagonal Image Processing with Evidential Probabilistic Supervised Classification Technique to Improve Image Retrieval Systems

Abstract

This paper presents a suggested approach to treat a major issue in images classification namely uncertainty. Uncertainty in image classification means some pixels within each cluster are more or less likely to actually belong to this cluster. So, techniques have been used in this paper to deal with the pixels that do not belong to specific regions, helping to raise image retrieval performance. This was done by merging one of the artificial intelligence techniques, which is image processing, with one of the statistical techniques for probability, which is evidential probabilistic. In such contexts, it may be advantageous to resort to two branches: hexagonal image processing based on partial down-sampling of the image resolution in both directions by half using weighted average performance then shifting the remaining pixels in alternate rows. The other is an evidential theory which is rich and flexible formalisms for representing and manipulating uncertain information. Both hexagonal image processing and evidential theory are used to obtain high accuracy in images classification. The hierarchical nature of the hexagonal image processing addressing scheme is exploited to extract features from the image efficiently.