Название: Quality Assessment of Visual Content
Автор: Ke Gu, Hongyan Liu, Chengxu Zhou
Издательство: Springer
Серия: Advances in Computer Vision and Pattern Recognition
Год: 2022
Страниц: 256
Язык: английский
Формат: pdf (true), epub
Размер: 45.9 MB

Image quality assessment (QA) is one of the basic techniques of image processing. It can evaluate the degree of image distortion by analyzing and studying the characteristics of images. In an image processing system, image QA plays an important role in system performance evaluation, algorithm analysis, and comparison.

This book provides readers with a comprehensive review of image quality assessment technology, particularly applications on screen content images, 3D-synthesized images, sonar images, enhanced images, light-field images, VR images, and super-resolution images. It covers topics containing structural variation analysis, sparse reference information, multiscale natural scene statistical analysis, task and visual perception, contour degradation measurement, spatial angular measurement, local and global assessment metrics, and more. All of the image quality assessment algorithms of this book have a high efficiency with better performance compared to other image quality assessment algorithms, and the performance of these approaches mentioned above can be demonstrated by the results of experiments on real-world images. On the basis of this, those interested in relevant fields can use the results obtained through these quality assessment algorithms for further image processing.

The single-image SR reconstruction upsamples an LR image to produce a high-quality HR image with finer details, which cannot be directly captured by a physical imaging system. The HR image obtained by HR cameras will result in a lot of production costs and manpower. In order to solve this problem, researchers have carried out extensive research on the SISR algorithm. SR image quality assessment technique as one of the most important parts of the SR technique field can evaluate the quality of SR images and the superiority of SR algorithms. Researchers have designed many SR QA methods and introduced Deep Learning (DL) techniques to better achieve objective QA of SR images. In the following content, we will introduce two Deep Learning-based SR image QA methods based on learning cascade regression and specific loss functions. By combining Deep Learning technology, the model can establish a more robust mapping relationship between the multiple natural statistical features and the visual perception scores.

The goal of this book is to facilitate the use of these image quality assessment algorithms by engineers and scientists from various disciplines, such as optics, electronics, math, photography techniques and computation techniques. The book can serve as a reference for graduate students who are interested in image quality assessment techniques, for front-line researchers practicing these methods, and for domain experts working in this area or conducting related application development.