Natural scene statistics-based blind visual quality assessment in the spatial domain

With the launch of networked handheld devices which can capture, store, compress, send and display a variety of audiovisual stimuli; high definition television (HDTV); streaming Internet protocol TV (IPTV) and websites such as Youtube, Facebook and Flickr etc., an enormous amount of visual data of visual data is making its way to consumers. Because of this, considerable time and resources are being expanded to ensure that the end user is presented with with a satisfactory quality of experience (QoE). While traditional QoE methods have focused on optimizing delivery networks with respect to throughput, buffer-lengths and capacity, perceptually optimized delivery of multimedia services is also fast gaining importance. This is especially timely given the explosive growth in (especially wireless) video traffic and expected shortfalls in bandwidth. These perceptual approaches attempt to deliver an optimized QoE to the end-user by utilizing objective measures of visual quality. In this thesis, we shall cover a variety of such algorithms that predict overall QoE of an image or a video, depending on the amount of information available for the algorithm design. Typically, quality assessment (QA) algorithms are classiffied on the basis of the amount of information that is available to the algorithm. This thesis will primarily focus on blind QA algorithms, where blind or no-reference (NR) QA refers to automatic quality assessment of an image/video using an algorithm which only utilizes the distorted image/video whose quality is being assessed. NR QA approaches are further classiffied on the basis of whether the algorithm had access to subjective/human opinion prior to deployment. Algorithms which use machine learning techniques along with human judgements of quality during the 'training' phase may be labelled 'opinion aware' algorithms. The first part of the thesis deals with such approaches. While such opinion aware-NR algorithms demonstrate good correlation with human perception on controlled databases, it is impossible to anticipate all of the different distortions that may occur in a practical system and hence train on them. In such cases, it is of interest to design QA algorithms that are not limited in their performance by training data. Approaches which operate without the knowledge of human judgements during the training phase are labelled as 'opinion unaware' (OU) algorithms. We propose such an approach in the second part of the thesis. Further, we propose new VQA algorithms in the last part of the dissertation to address the completely blind VQA problem. The proposed approach quantify disturbances introduced due to distortions and thereby predict the quality of distorted content even without any external knowledge about the pristine natural sources and hence zero shot models.