Interactive Image Recommended

With the rapid development of Internet and multimedia technology,the number of images and video resources is explosively expanding,how to find the target image quickly and efficiently which users search for in vast amounts of resources,and how to feedback the results to users effectively by the retrieval system.It's no doubt an urgent problem to be solved.The emergence and development of the content-based and semantic-based image retrieval system was realized in this context. However, the existing techniques exists the following problems:first, it exists a semantic gap between automatic image annotation and the real meaning of the image; second, for the same picture, there may be a different image annotation; third, Users often can not find the exact keywords to describe the target image that they want to search for; fourth, the search UI is too simple that is not easy enough for the user to view the retrieve result and search again.Moreover,which lacks of convenient feedback mechanism for interaction.For the above reasons, in this aticle we have done some related reserch and further study,proposed a new dynamic interactive image recommendation technology.Interactive Image recommendation technology is developed based on the content-based and the semantic-based image retrieval techniques, it is a new image recommendation techniques based on keywords. This paper analyzed semantic relations between the concept of the image annotation, as well as the potential probability relationship to build the concept semantic network. At the same time, it extract color feature, texture features of the image visual content, and integrate these two basic visual features into a mixed feature to represent the single image visual features. On this basis of these, we have fully integrated the low-level visual features and the high-level semantic features to search. In order to show more results in a limited two-dimensional plane.In this paper we introduced the hyperbolic visualization technology into the field of image retrieval, proposed a new image visualization model, named the image display model hyperbolic space.Firstly, Interactive image recommendation system establishes image semantic network through the concept of annotation,then extracts the visual features and clusters the image,displays in hyperbolic. Sencondly,users select the target image class by observing to feedback the retrieve.After receiving the feedback message,according to the visual image characteristics and the feedback message,the system will go to re-sample automatically,recommend effectively target category to show in the two-dimensional plane,and complete the process of image selection feedback dynamic interaction.Finally, we have analyzed and done experiment on LabelMe and Corel image library, achieved good experimental results and proved that this method is effective.