Stereoscopic Image Quality Assessment

verfasst von: Prof. Yong Ding, Guangming Sun

Verlag: Springer Singapore

Buchreihe : Advanced Topics in Science and Technology in China

Enthalten in: Springer Professional "Wirtschaft+Technik" , Springer Professional "Technik" , Springer Professional "Wirtschaft"

Einloggen, um Zugang zu erhalten

Über dieses Buch

This book provides a comprehensive review of all aspects relating to visual quality assessment for stereoscopic images, including statistical mathematics, stereo vision and deep learning. It covers the fundamentals of stereoscopic image quality assessment (SIQA), the relevant engineering problems and research significance, and also offers an overview of the significant advances in visual quality assessment for stereoscopic images, discussing and analyzing the current state-of-the-art in SIQA algorithms, the latest challenges and research directions as well as novel models and paradigms. In addition, a large number of vivid figures and formulas help readers gain a deeper understanding of the foundation and new applications of objective stereoscopic image quality assessment technologies.

Reviewing the latest advances, challenges and trends in stereoscopic image quality assessment, this book is a valuable resource for researchers, engineers and graduate students working in related fields, including imaging, displaying and image processing, especially those interested in SIQA research.

Inhaltsverzeichnis

Frontmatter

Chapter 1. Introduction

Abstract

Nowadays, objective image quality assessment (IQA) plays an important role for performance evaluation of image/video processing systems. Over the past few years, a variety of IQA methods have been introduced and they can be divided into three categories: full-reference IQA, reduced-reference IQA and no-reference IQA. All of these methods are clarified in detail in this book. In this chapter, the overall structure of the book is explained briefly and a summary of each of the following chapters is also provided.

Yong Ding, Guangming Sun

Chapter 2. Brief Introduction on 2D Image Quality Assessment

Abstract

In this chapter, a brief introduction about 2D image quality assessment is given. Firstly, some public image quality databases are introduced which provide ground-truth information for training, testing and benchmarking. Secondly, IQA performance metrics including SROCC, KROCC, PLCC and RMSE to compare the accuracy of different IQA methods are provided. Finally, the general frameworks of 2D IQA methods containing full-reference (FR), reduced-reference (RR) and no-reference (NR) are illustrated based on specific algorithms.

Yong Ding, Guangming Sun

Chapter 3. Difference Between 2D and Stereoscopic Image Quality Assessment

Abstract

With the rapid development of stereo and multi-view systems and a wide adoption of these systems, stereoscopic image quality assessment (SIQA) has become an important and challenging problem faced in numerous application such as 3D films, stereo visualization and 3D enhancement. Compared with 2D image quality assessment methods, SIQA needs considering complex binocular visual properties. The difference between SIQA and 2D IQA is introduced firstly in this chapter. Then stereoscopic image quality databases are listed and discussed detailedly. Finally, the general designed frameworks of SIQA methods are provided.

Yong Ding, Guangming Sun

Chapter 4. SIQA Based on 2D IQA Weighting Strategy

Abstract

As image quality assessment (IQA) methods for plant images have been explored thoroughly, some SIQA algorithms apply 2D IQA methods on both stereoscopic views independently and then combine the two scores to obtain an overall quality score by a dedicated strategy. The early algorithms only combine the two scores simply to obtain the final quality score, and some improved algorithms utilize both the stereoscopic views and the depth/disparity information. All of these algorithms could achieve fairly good performance. The mainstream and state-of-art SIQA based 2D IQA weighting strategy are introduced in detail in this chapter.

Yong Ding, Guangming Sun

Chapter 5. Stereoscopic Image Quality Assessment Based on Binocular Combination

Abstract

Only employing depth information in stereoscopic image quality assessment models cannot simulate human visual characteristics well. Thus, a cyclopean image generated from the left and right views is designed to overcome this defect. Different methods to generate a cyclopean image are discussed firstly in this chapter. Then two region classification strategies to deal with no-matched pixels caused by different angles of two views are introduced. Finally, visual fatigue and visual discomfort prediction models are developed to simulate the negative influence of non-corresponding areas in stereo pairs.

Yong Ding, Guangming Sun

Chapter 6. Stereoscopic Image Quality Assessment Based on Human Visual System Properties

Abstract

Modelling the behavior of Human Visual System (HVS) is the ultimate target of Image Quality Assessment (IQA). The hierarchical structure of HVS and different HVS models are introduced firstly in this chapter. And some classical IQA methods based on the hierarchical structure of HVS are discussed in detail. Visual attention, as one of the most important mechanisms of the HVS, is clarified clearly and some Stereoscopic Image Quality Assessments (SIQA) methods based on visual saliency are also presented. In the end of this chapter, Just Noticeable Difference (JND) model and corresponding IQA methods are introduced.

Yong Ding, Guangming Sun

Chapter 7. Stereoscopic Image Quality Assessment Based on Deep Convolutional Neural Models

Abstract

The deep convolutional neural network (CNN) has achieved great success in image process areas in recent years. Many image quality assessment methods directly use CNN for quality prediction. Optimizing deep convolutional neural network with high generalization ability needs a huge amount of data, however, the most popular IQA databases are usually too small. Therefore, transfer learning and patch-wise strategy are developed to realize data enhancement. On the basis of alleviating the insufficient training data, some methods improve the CNN framework to better simulate HVS, and the implementation details are described in this chapter. Finally, some necessary related knowledges about CNN-based IQA methods are introduced.

Yong Ding, Guangming Sun

Chapter 8. Challenging Issues and Future Work

Abstract

The complete development history of Stereoscopic Image Quality Assessment (SIQA) has been overviewed in the previous chapters. Even if the state-of-art SIQA methods have achieved competitive results, there still have some challenges and obstacles that need to be discussed and concluded in the end of this book, including subjective studying, in-depth research of Human Visual System (HVS) and the bottleneck of insufficient training data. Finally, the practical applications of SIQA are discussed.

Yong Ding, Guangming Sun

Titel: Stereoscopic Image Quality Assessment
verfasst von: Prof. Yong Ding
Guangming Sun
Verlag: Springer Singapore
Electronic ISBN: 978-981-15-7764-2
Print ISBN: 978-981-15-7763-5
DOI: https://doi.org/10.1007/978-981-15-7764-2