Image cultural analytics through feature-based image exploration and extraction

Abstract

Today, we are witnessing the proliferation of digital media, whether through mass digitization efforts, or through the publishing activities of the multitude of social-platform users. Much of this data is not the product of formal work'', but rather the product of artistic and social activity and the nascent field of cultural data mining'' focuses on methods for analyzing this data to extract and visualize interesting patterns that give us insights of how our culture evolves. This field adopts and combines methods from image processing, data mining, probability and statistics, and visualization to extract and represent information about culture. In this research, we report on a image feature-based approach to the cultural-analytics problem, whereby we analyze images in terms of a set of low-level and mid-level features, \ie colors, textures, and shapes, and examine the correlations of these features with high-level semantic information of cultural significance. In our current implementation, using a faceted search interface, a combination of one or more visual features or tags can be used to perform image- and/or tag-based queries on an image repository. Our results on some sample data sets, obtained from publicly available sources, show that aggregated image features can be used for inferring higher-level cultural information.