Fig. 3
From: ConIQA: A deep learning method for perceptual image quality assessment with limited data
Comparing the slope and intercept of the fitted trend line for predictions by different metrics as a function of ground truth MOS values. The linear trend is defined as \(MOS'=aMOS+b\) where a, and b are slope, and intercept respectively. (a) The scatter plot compares the linearity of predictions by ConIQA and IW-SSIM as a function of ground truth MOS values. (b) Pearson correlation coefficient vs. slope of the fitted lines for all metrics. The dotted black lines represent the ideal values for the correlation coefficient, \(r=1\), and the slope, \(a=1\). (c) Intercept and slope of the fitted trend lines. The dotted black lines represent the ideal values for the correlation coefficient, \(r=1\), and the intercept, \(b=0\).
