Seminar: Blind Image Quality Assessment

Hao Cai
Ph.D. Oral Comprehensive

Supervisory Committee: Drs. Minglun Gong, Wolfgang Banzhaf and Lourdes Pena-Castillo

Blind Image Quality Assessment

Department of Computer Science
Wednesday, December 19, 2018, 2:30p.m., Room EN 2022


Abstract

Image quality assessment (IQA) plays an important role in numerous image/video processing and computer vision applications, including image compression, image transmission and image restoration. The goal of objective IQA is to develop a computational model that can predict image quality in a way that is consistent with human perception. Compared with subjective quality evaluations (i.e., psycho-visual tests), objective IQAmetrics can automatically and accurately assess the perceptual quality of digital images in a timely manner.

The proposed research focuses on blind IQA (BIQA), which predicts image quality without access to reference images. First, a BIQA metric for gamut-mapped images is introduced. Considering both the local and global aspects of distortions in gamut-mapped images, two categories of natural scene statistics are analyzed. Experimental results on three gamut mapping databases demonstrate that our method outperforms the relevant state-of-the-arts. To further validate its effectiveness, the proposed metric is applied for bench marking gamut mapping algorithms. Secondly, a general-purpose BIQA method is presented, which can evaluate the quality of digital images without prior knowledge on the types of distortions. The proposed metric investigates second-order statistics in both the wavelet domain and spatial domain on natural images. Experimental results on several publicly available image quality databases demonstrate the superiority of our approach.

Most of the current BIQA methods quantify quality degradations based on natural scene statistics, where the frameworks usually involve separate feature extraction stage and quality prediction stage. It is worth noting that the recent exciting development of deep neural network (DNN) methodologies have shown great promises in many vision tasks. As future work, I plan to develop another general- purpose BIQA model based on DNN, which aims at providing an end-to-end solution to quality assessment.