Time and Location:   March 31, 2004, 2-3:00pm, Room 1314.

Title:     Perceptual image quality assessment: from error visibility to structural similarity

Speaker:     Zhou Wang, Laboratory for Computational Vision, NYU


Abstract:

Image quality measures that can  automatically predict perceived image
quality play important roles in various image processing applications.
Traditional methods attempt to quantify the visibility of errors
(differences) between a distorted image and a reference image using a
variety of known properties of the human visual system. We discuss the
limitations and drawbacks of this framework. Under the assumption that human
visual perception is highly adapted for extracting structural information
from a scene, we introduce an alternative complementary framework for
quality assessment based on structural similarity. We demonstrate the
promise of our method through a set of intuitive examples, as well as
comparisons to both subjective ratings and state-of-the-art objective
methods on a large image database. Furthermore, we propose a new methodology
for the evaluation and comparison of image quality measures, in which we
synthesize image stimuli that best differentiate two candidate quality
measures. This process results in striking examples indicating the relative
strength and weaknesses of the quality measures being compared, and also
suggests potential improvement of these measures.

Papers, demonstrations and Matlab code can be found at
http://www.cns.nyu.edu/~lcv/ssim/