Automatic clustering for digital photograph collections using time and content information
Wan Hyun Cho, In Seop Na, Soo-Hyung Kim, Jun Yong Choi, Jong Hyun Park, Dung Phan · 2013
In this paper, we propose an automatic clustering method for large photographs collections using time and content features. First, we think about various types of feature vectors that are suitable to represent time and content of photographs, and we computed the similarity measures that can be represented an affinity between these photos. Next, we consider a clustering method for photo collection. Here, we first build a coarser clustering by automatically partitioning a given photo collection into several clusters using the Mean shift clustering algorithm. Second, we construct dense clustering by optimizing a Gaussian Dirichlet process mixture model taking initial clusters model as coarser clustering result. Finally, we have conducted the experiment which is able to evaluate a performance of our clustering method for various events photos collection. Experimental results show that both three types of features and Gaussian Dirichlet process mixture model brings about higher values of accuracy and precision in the clustering of photo-collection.