The field of image and video quality assessment (QA) addresses the question of how to quantitatively model user experience, and use these models to predict visual quality in accordance with visual perception. QA models are useful in providing benchmarks on performance, leading to improved design and control of image and video systems. In this talk, we discuss reduced reference visual quality assessment, where only partial information about the reference is available for quality assessment. Using natural scene statistical models, we design image QA algorithms that measure the information change between the reference and distorted images through entropic differences. This leads to a family of algorithms that vary in the amount of reference information required for QA, ranging from almost no information to full information from the reference. We extend this framework to videos by developing statistical models along the temporal dimension and computing temporal entropic differences. Our image and video QA algorithms are shown to correlate very well with human perception of quality on large publicly available databases.
Rajiv Soundararajan received the B.E.(Hons.) degree in Electrical and Electronics Engineering from Birla Institute of Technology and Science (BITS), Pilani in 2006 and the M. S. and Ph. D. degrees in Electrical and Computer Engineering from The University of Texas at Austin in 2008 and 2012 respectively. He has been with Qualcomm Research India, Bangalore since 2012. His research interests are broadly in statistical image and video processing, information theory, machine learning and computer vision.