LINIQE（一）

论文标题：
A Feature-Enriched Completely Blind Image Quality Evaluator
一种特征被强化的完全无参考图像质量评价方法

依据训练模型中是否with associated human subjective scores，可以将BIQA方法分为：
- opinion-aware：
weak generalization capability（样本有限失真类型、少混合失真图片）；
dependent to databases（得到模型依赖选取数据库，难以实际运用）；
- opinion-unaware；

本文是介绍的是一种基于opinion-aware方法NIQE改进的opinion-unaware方法。同时相对NIQE，此法：
- two types of NSS features +introduce three additional types of quality-aware features
- a single global MVG model → fit the feature vector to an MVG model

方法摘要概述：
——integrating the features of natural image statistics derived from multiplecues
——learn a multivariate Gaussian model of image patches from a collection of pristine natural images.
—— a Bhattacharyya-like distance is used to measure the quality of each image patch
——an overall quality score is obtained by average pooling
整合多重线索提取的自然图像统计特征，利用收集的大量原始自然图形的分块训练多元高斯模型，用类巴氏举例在模型中判断每块图像质量，平均池策略得到整体质量。

提取的五类NSS统计特征：
- locally mean subtracted and contrast normalized (MSCN) coefficients （去局部均值差别标准化的系数分布）
- the distribution of products of pairs of adjacent MSCN coefficients （相邻系数两和产生的分布？）
- gradient features (see Sect. II-C).
- use log-Gabor filters extract statistical features from the filter responses (see Sect. II-D).
- Color distortions (see Sect. II-E)在对数衡量的对立彩色空间中，根据图像强度分布得到的统计特征 .