October 5, 2011
Title: Learning a Blind Measure for Perceptual Image Quality
Abstract: In this talk I present our paper on blind measurement of perceptual image quality in CVPR2010. Evaluating the perceptual quality of an image is a desirable but challenging problem. However, most commonly used measure do not map well to human judgement of image quality. A further complication of many existing image measure is that they require a reference image, which is often not available in practice. We propose a "blind" image quality measure, where potentially neither the ground-truth image nor the degradation process are known. Our method uses a set of novel low-level image features in a machine learning framework to learn a mapping from these features to subjective image quality scores. The image quality features stem from natural image measure and texture statistics. Neighbourhood embedding of our proposed features well clusters images of similar quality and relevant distortion type, indicating good potential to generalize our learned measure to new images and distortion types. Experiments on a standard image quality benchmark dataset shows that our method significantly outperforms state of art no-reference image assessment algorithm in all aspects.
Biography: Huixuan Tang is a PhD student at University of Toronto. Her research interest lies in computer vision and machine learning, with an concentration in computational photography. During her study, she interned at Microsoft Research Asia in 2006-2007 and Microsoft Research in 2010.