A Survey on Learning Crowdsourced User Preferences for Visual Summarization of Image Collections

We present a new approach for selecting images which are suitable for inclusion in the visual summaries. This approach is designed on the basis of how people generally think of summarizing image collections. For obtaining large number of manually created visual image summaries and criteria which guide users for selection of images we use Amazon Mechanical Turk Crowd sourcing platform. This technique utilize the content and context of images, image popularities, similarities between images, sentimental analysis. We describe images not only on the basis of their properties but also we consider the fact behind images that are related semantically. This increases efficiency and enables aesthetic appeal, proliferation of sentiment, and various emotions associated with a particular group of images. We examine the trend of a low inter-user contract, which is helping to make a computerized evaluation of aesthetic summaries and propose a solution influenced by the text summarization and machine interpretation communities. The studies conducted on a collection of geo-referenced Flickr image collections demonstrate the potency of our image selection approach.