無參考圖像質量評價算法(NIQE):Making a Completely Blind Image Quality Analyzer

paper:Making a Completely Blind Image Quality Analyzer
date:2012
author:Anish Mittal etc...
code:NIQE

1.Introduction

本文提出的方法Natural Image Quality Evaluator(NIQE)是沿襲brisque的，但是不需要在人類評分的數據集上訓練，所以是"opinion unaware"，"completely blind"的。

2.NIQE

A.Spatial Domain NSS

類似BRISQUE方法，提取NSS特征。對於圖像\(I(i,j)\),經正則化得到mean subtracted contrast normalized(MSCN)系數

\[\hat{I}(i,j)=\frac{I(i,j)-\mu(i,j)}{\sigma(i,j)+C} \]

其中,\(i \in 1,2,\dots,M,j\in 1,2,\dots,N\),\(M,N\)是圖像的高和寬，\(C=1\)是為了數值穩定的常數。

\[\mu(i,j)=\sum_{k=-K}^{K}\sum_{l=-L}^{L} \omega_{k,l}I_{k,l}(i,j)\\ \sigma(i,j)=\sqrt{\sum_{k=-K}^{K}\sum_{l=-L}^{L}\omega_{k,l}(I_{k,l}(i,j)-\mu(i,j))^2} \]

其中\(\omega=\{ \omega_{k,l}\vert k=-K,\dots,K,l=-L,\dots,L\}\)是高斯核。

B.Patch Selection

Since humans appear to more heavily weight their judgments of image quality from the sharp image regions, more salient quality measurements can be made from sharp patches.

由於人的主觀評價對於圖像銳利區域更為敏感，所以先提取銳利區域再計算特征。將圖像划分為\(P*P\)個patches,利用下面規則選取銳利部分:

\[\delta(b)=\sum\sum_{(i,j)\in patch_b}\sigma(i,j) \quad b=1,2,\dots,P*P　　\tag{1} \]

設定閾值T,如果\(\delta(b)>T\),認為\(patch_b\)是銳利的，選擇這些patches作為計算特征的部分。

C.Characterizing Image Patches

上面經過sharpness criterion(1)選擇一些patches之后，需要提取特征，類似於BRISQUE方法擬合GGD和AGGD得到18維特征，在不同尺度下再做一次，得到36維特征。

D.Multivatiate Gaussian Model(MVG)

MVG

\[f_X(x_1,\dots,x_k)=\frac{1}{(2\pi)^{k/2}\vert\Sigma\vert^{1/2}}exp(-\frac{1}{2}(x-\nu)^T\Sigma^{-1}(x-\nu)) \]

其中\((x_1,\dots,x_k)\)是上面計算出的36維NSS特征。\(\nu,\Sigma\)表示多元高斯分布的均值和協方差，作者用125張natural images擬合出\(\nu,\Sigma\)的值。

E.NIQE Index

NIQE分數的計算，是通過計算待測圖片MVG模型參數和上面得到的自然圖片MVG模型參數的距離來得到(如下式)。不過選擇patch的准則(1)不應用到待測圖片上，而只用在上面自然圖片模型參數估計上。原因如下:

The sharpness criterion (1) is not applied to these patches because loss of sharpness in distorted images is indicative of distortion and neglecting them would lead to incorrect evaluation of the distortion severity.

\[D(\nu_1,\nu_2,\Sigma_1,\Sigma_2)=\sqrt{(\nu_1-\nu_2)^T(\frac{\Sigma_1+\Sigma_2}{2})^{-1}(\nu_1-\nu_2)} \]

3.PERFORMANCE EVALUATION

測試結果: