Image Retrieval Approach Based On Attention Driven Model

This dissertation discusses some key problems on content-based image retrievaltechniques focusing on shortcomings of current systems in image interpretation. An imageretrieval scheme based on visual attention driven model is proposed, emphasizing on imagesegmentation andtheextractionofsalient objects.Based onthediscussionabove,anintegrityimage retrievalsystemisdescribed,anditseffectivenessisprovedintheexperiments.Problem is that most of content based image retrieval systems are based onglobal/semi-global featuresorbasedon regions features.This facilitates fast retrieval but leadto decrease the rate of retrieval, because there are some other regions irrelevant within animage. A key problem to be solved is that which region is more attentive than others in animage, or which region is more informative or interesting to a human being. This problem isbasically unsolved because it includes the understanding and interpretation of an image.Fortunately, several computational models of visual attention inspired biologicallyare helpfulfor machine analysis and understanding of image. In this dissertation, a novel content-basedimage retrieval approach using attention-driven model is proposed. In this approach, firstly,images are segmented by EM algorithm, then saliency map are generated by Itti-Koch visualattentionmodel,atlastsalientregionsaregotbyregiongrowingalgorithm.A novel content-based image retrieval approach usingattention-driven model isproposedin this dissertation. It imitates the attention process of human for processing images,interpreting an image based on several attentive objects and the background. The regionsextracted from image are more attentive. This attention-driven modelused for content-basedimage retrieval provides a promising performance, as compared with some current peersystemsintheliterature.