Finding Interesting Images in Albums using Attention

Abstract

Commercial systems such as Flickr display interesting photos from their collection as an interaction mechanism for sampling the collection. It purely relies on social activity analysis for determining the notion of interestingness. We propose an alternative technique based on content analysis for finding interesting photos in a collection. We use a combination of visual attention models and an interactive feedback mechanism to compute interestingness. A differentiating feature of our approach is the ability to customize the set of interesting photos depending upon the individual interest. Also, we incorporate non-identical duplicate detection as a mechanism to strengthen the surprise factor among the potentially interesting set of candidate photos. We have implemented the system and conducted a user study whose results are promising. This proposed work presents a variant on query by example integrating user relevance feedback to choose “interesting” photos.