Assessing the Quality of User―Generated Content.docVIP

Assessing the Quality of User―Generated Content 　　Abstract 　　With the widespread use of digital cameras， imaging software， photo-sharing sites， social networks， and other related technologies， media production and consumption patterns have become much more multifaceted and complex than they used to be. User-generated content in particular has grown tremendously. As a result， quality of experience （QoE） and related quality assessment （QA） methods must also be looked at from a different angle. This paper contrasts some of the traditional quality assessment approaches with newer approaches designed for user-generated content. It also describes some sample applications we have developed. 　　Keywords 　　image quality； photo collections； social media； photowork； summarization 　　1 Introduction 　　Quality assessment （QA） for images dates back to the 1970s， when the first studies were done on the visual cortex， vision modeling， and digital imaging. Algorithms based on models of the human visual system now compete with more pragmatic， image- or feature-based methods [1]. Video QA has a similar， albeit shorter， history [2]. 　　Ubiquitous and affordable digital cameras （also in the form of embedded image capture devices） now enable users to take pictures and videos almost anywhere， anytime. This has led to an explosion in the amount of picture material produced by both amateurs and professionals. 　　However， traditional media and user-generated content are fundamentally different on many levels， especially from a quality assessment perspective （Table 1）. 　　1.1 QA for Traditional Media 　　Traditional QA methods have focused mainly on the processing and distribution chains of broadcast media， that is， the compression， transmission， and enhancement of images and video. In many cases， there is an explicit reference （e.g. the source image or video）， that passes through the system and undergoes certain changes （e.g. loss of fidelity， compression artifacts， packet loss， noise