ABSTRACT
The term “image cultural analytics” refers to the study of image data, generated from social and cultural activities, for the purpose of gaining insight on how our culture evolves. As an increasing body of image data becomes available, either born digital or digitized, such study can substantially benefit from computational methods. In this paper, we report on our work in this area, which investigates the correlations between automatically extracted, low-and mid-level image features with semantic information, captured by image metadata. Our work makes two contributions. First we have developed a prototype tool for exploring correlations of interest in image repositories, through a faceted-search interface that supports users to combine visual features and metadata. Second, we have demonstrated the potential of this method by conducting an empirical study with the IMDB dataset.
