No Reference Image Quality Assessment Method for Blurred Image

摘要

Considering that many no reference image quality assessment methods cannot give better assessment results for blurred images, this study proposes a no-reference image quality assessment method which has better assessment results. The method firstly extracts texture and structure features of a blurred image and then calculates its texture similarity and structural similarity. Finally, with these texture similarity and structural similarity as the input factors, the subjective assessment value DMOS provided by LIVE database as the output factor, a [2 9 1] Back-Propagation (BP) neural network prediction model is built. The experimental results show that the prediction model is stable and the differences between subjective assessment values and prediction results are small. Their Correlation Coefficients are all above 0.9.