Finding interesting photos in albums using visual attention

Abstract

Finding the interesting photos from albums and e-chronicles is new problem in the multimedia community nowadays. Interestingness varies from person to person due to their individual interest, context, experiences and preferences. One way of capturing user interest is studying how human beings pay attention towards various objects. The other way could be asking user what seems interesting to her due to the variation in interpretation and accumulation of experience. The former can be achieved by visual attention model and later by a relevant feedback mechanism. In this thesis, we propose a framework to address the problem of fetching interesting images by using visual attention model and pseudo relevant feedback mechanism. We implemented a web based system prototype ``EChronicles Attention System" based on the framework. We conducted a user study to investigate whether content based interestingness based on attention model coupled with relevance feedback would be useful or not. The results are encouraging. Also, we substantiate the utility of non identical duplicate detection in increasing the surprise factor for the user.