Mining salient images from a large-scale blogosphere

User-generated images are now prevalent across social media platforms, such as Facebook, Twitter, and various blogospheres. These images can be categorized and ranked based on their relevant topics. In this paper, we present and compare candidate schemes for mining salient images related to a specific topic or object among a large number of images from a blogosphere. Identifying salient images consists of several steps: calculating the similarity between images, k-means clustering images, and ranking images. In each step, we propose a set of alternatives and as a result, present an optimal combination scheme by conducting an empirical comparison of the performance of each scheme. Furthermore, to address scalability, we also present a distributed version of the schemes and experimental results based on MapReduce on top of a Hadoop environment.