1.2 Image Quality Assessment Models 3
IQA models are usually classified into three families (see (Lahoudou et al. 2011;
Beghdadi et al. 2013) for a brief review of IQA and machine learning):
• Full-reference models that use the original version of the image for estimating
the quality of the processed version. These models are the most used methods to
evaluate image quality (for example, the well-known PSNR and SSIM). They are
easy to compute in real time and correlated with human subjective appreciation
but require a reference image. Unfortunately, these models are not applicable for
computer-generated images since the final reference image is not already known
during the image generation process.
• No-reference models that evaluate the quality of images without access to reference
images. Some recent papers (Ferzli and Karam 2005; Zhang et al. 2011) proposed
no-reference quality assessment methods with good results but limited to JPEG
images. Other methods were proposed for computer-generated images with some
success (Delepoulle et al. 2012) but a complete framework has to be yet defined.
• Reduced-reference models that analyze the processed image using some relevant
information to calculate the quality of the result image. This model seems to be
particularly interesting for our study as we will show in the following of the book.
In the last decade, numerous IQA methods for computer-generated images have
been proposed but the resulting models are limited in practice and they are still under
investigation. Currently, the classical model to characterize image quality remains
psycho-visual experiments (Human in the loop experiment (Faugeras 1979)).
1.3 Organization of the Book
In this book, we assume that the reader is familiar with the basic aspects of machine
learning and image processing, and we only focus on the visual quality assessment
of computer-generated images using soft computing. We present recent techniques
to assess if such a photo-realistic computer-generated image is noisy or not, based
on full-reference, reduced-reference, and no-reference image quality methods, using
learning machines and fuzzy sets. These techniques make it possible to set up uncer-
tainty brought by perceptual noise affecting the image synthesis process. Note that
we mainly focus on grayscale images, or more precisely the “L” component of “Lab”
color images, since noise mainly affects this component, ( Carnet et al. 2008).
In Chap. 2, we introduce image rendering to the reader. We present the basic
notions of light transport and the equations that formalize rendering. We then recall
the Monte Carlo method and detail the path-tracing algorithm which is the core of
many renderers currently used in the computer graphics industry.
剩余95页未读，继续阅读
wang1062807258
- 粉丝: 13
- 资源: 272
我的内容管理
展开
最新资源
- AirKiss技术详解：无线传递信息与智能家居连接
- Hibernate主键生成策略详解
- 操作系统实验：位示图法管理磁盘空闲空间
- JSON详解：数据交换的主流格式
- Win7安装Ubuntu双系统详细指南
- FPGA内部结构与工作原理探索
- 信用评分模型解析：WOE、IV与ROC
- 使用LVS+Keepalived构建高可用负载均衡集群
- 微信小程序驱动餐饮与服装业创新转型：便捷管理与低成本优势
- 机器学习入门指南：从基础到进阶
- 解决Win7 IIS配置错误500.22与0x80070032
- SQL-DFS：优化HDFS小文件存储的解决方案
- Hadoop、Hbase、Spark环境部署与主机配置详解
- Kisso：加密会话Cookie实现的单点登录SSO
- OpenCV读取与拼接多幅图像教程
- QT实战：轻松生成与解析JSON数据
资源上传下载、课程学习等过程中有任何疑问或建议，欢迎提出宝贵意见哦~我们会及时处理！
点击此处反馈
