A summarized photo visualization system with maximal clique finding algorithm

Abstract

The affordability of digital cameras, storage, processors and the advances made in these areas are encouraging people to continuously take hundreds of photos. However, managing the large number of photographs involves arduous tasks such as selecting good quality photos and classifying and labeling each photo. Generally, users put their photos into certain user-designated folders on their local PCs without considering any classified information. One of the main problems related to this management method is that users do not systematically create their photo folders because they are careless and apathetic. This practice results in confusion when the users want to find their photos. One method to overcome this problem is to construct a central photo management system that can manage many photos on the user's local PC. This paper proposes an integrated photo management system coupled with a database on the web, which provides users with an automated photo clustering and visualization function that allows photo overlaps. The proposed system provides spatial clustering for Nearly Identical Photos, and it places photos with overlaps to improve space efficiency for the user. This system also provides users with a CUDA version of Depth of Field evaluation and blur estimation functions. In order to evaluate our system, we conducted two quantitative experiments relating to space efficiency and clustering correctness. First, we investigate the placed photo areas of ACDSee (grid layout) and our system to evaluate how much screen space is saved by nearly identical photos overlapping. Second, we also calculate the precision and recall of our system and Cooper's with regard to user-classified photo sets.