Blind Image Quality Assessment
Duration: August 2014 - April 2015
Status: Completed
Members: Pranav Sodhani, Akshat Bordia, Dr. Kannan Karthik
Overview
Blind Image Quality Assessment or No-Reference Image Quality Assessment (IQA) refers to evaluating the quality of a (possibly) distorted image in the absence of original reference image. The trained human eye does not need a reference to gauge the quality of a specific image or photograph presented to it. This opens us the question - "Can a signal processing unit rate the quality of a particular image?". The process has to be content-independent and must mimic human perception. Given the enormous variability of content in images and presence of different types of distortion, Blind IQA is a challenging research problem.
We approach the problem by developing Blind IQA metrics for blur and noise individually, knowing that these are the two most prominent forms of distortion. Our work proposes two metrics namely, CHARM (for blur) and CINEMA (for noise). Experimental results on existing databases show that both CHARM and CINEMA correlate well with human perception and perform at-par with the top NR IQA algorithms. Interested readers might check the following website to read more on CHARM and CINEMA.
Webpage: https://sites.google.com/site/blindiqa/home
The thesis can been downloaded by clicking HERE. Our work led to the following publications:
Publications
[1] (Status: Accepted) - Pranav Sodhani, Akshat Bordia and Dr. Kannan Karthik, "Blind Content Independent Noise Estimation for Multimedia Applications", ICISP 2015, Bangalore, India.
[2] (Status: Accepted) - Pranav Sodhani and Dr. Kannan Karthik, "No Reference Blurred Image Quality Assessment in the Spatial Domain", ICIG 2015, Tianjin, China
Poster Presentation