Image Quality Assessment of Computer-generated Images: Based on Machine Learning and Soft Computing (SpringerBriefs in Computer Science)

✍ Scribed by André Bigand, Julien Dehos, Christophe Renaud, Joseph Constantin

• Publisher: Springer
• Year: 2018
• Tongue: English
• Leaves: 96
• Category: Library

✦ Synopsis

Image Quality Assessment is well-known for measuring the perceived image degradation of natural scene images but is still an emerging topic for computer-generated images. This book addresses this problem and presents recent advances based on soft computing. It is aimed at students, practitioners and researchers in the field of image processing and related areas such as computer graphics and visualization.

In this book, we first clarify the differences between natural scene images and computer-generated images, and address the problem of Image Quality Assessment (IQA) by focusing on the visual perception of noise. Rather than using known perceptual models, we first investigate the use of soft computing approaches, classically used in Artificial Intelligence, as full-reference and reduced-reference metrics. Thus, by creating Learning Machines, such as SVMs and RVMs, we can assess the perceptual quality of a computer-generated image. We also investigate the use of interval-valued fuzzy sets as a no-reference metric.

These approaches are treated both theoretically and practically, for the complete process of IQA. The learning step is performed using a database built from experiments with human users and the resulting models can be used for any image computed with a stochastic rendering algorithm. This can be useful for detecting the visual convergence of the different parts of an image during the rendering process, and thus to optimize the computation. These models can also be extended to other applications that handle complex models, in the fields of signal processing and image processing.

✦ Table of Contents

Preface
Acknowledgements
Contents
About the Authors
1 Introduction
1.1 Natural-Scene Images, Computer-generated Images
1.2 Image Quality Assessment Models
1.3 Organization of the Book
References
2 Monte Carlo Methods for Image Synthesis
2.1 Introduction
2.2 Light Transport
2.2.1 Radiometry
2.2.2 Formulation of Light Transport
2.3 Monte Carlo Integration
2.3.1 Monte Carlo Estimator
2.3.2 Convergence Rate
2.3.3 Variance Reduction Using Importance Sampling
2.4 Path Tracing
2.4.1 Random Walk
2.4.2 The Path-Tracing Algorithm
2.4.3 Global Illumination
2.5 Conclusion
References
3 Visual Impact of Rendering on Image Quality
3.1 Introduction
3.2 Influence of Rendering Parameters
3.2.1 Path Length
3.2.2 Number of Path Samples
3.3 Influence of the Scene
3.3.1 Light Sources
3.3.2 Scene Geometry
3.3.3 Materials
3.4 Conclusion
References
4 Full-Reference Methods and Machine Learning
4.1 Image Quality Metrics Using Machine Learning Methods
4.2 Experimental Setup
4.2.1 Overview
4.2.2 Data Acquisition
4.2.3 Psycho-visual Scores Acquisition
4.3 Noise Features Extraction
4.3.1 Classical Strategies
4.3.2 Pooling Strategies and Deep Learning Process
4.4 Image Quality Metrics Based on Supervised Learning Machine
4.4.1 Support Vector Machines
4.4.2 Relevance Vector Machines
4.5 Conclusion
References
5 Reduced-Reference Methods
5.1 Introduction
5.2 Fast Relevance Vector Machine
5.3 Image Quality Evaluation (IQE)
5.3.1 IQE Using FRVM
5.3.2 IQE Using Inductive Learning
5.4 Experimental Results and Discussion
5.4.1 Design of the Inductive Model Noise Features Vector
5.4.2 Inductive SVM Model Selection
5.4.3 Experiments Using Inductive Learning
5.4.4 Comparison with the Fast Relevance Vector Machine
5.5 Conclusion
References
6 No-Reference Methods and Fuzzy Sets
6.1 Introduction
6.2 Interval-Valued Fuzzy Sets
6.2.1 Uncertainty Representation
6.2.2 IVFS Entropy
6.3 IVFS for Image Noise Estimation
6.3.1 Design of the IVFS Image Noise Estimation
6.3.2 Proposed Scheme
6.3.3 Algorithm and Implementation
6.4 Experimental Results with a Computer-generated Image
6.4.1 Image Database for Noise Estimation
6.4.2 Performances of the Proposed Method
6.5 Conclusion
References
7 General Conclusion and Perspectives
7.1 Summary
7.2 Perspectives