Open AccessThree Different Features Based Metric To Assess Image Quality Blindly
Author(s) -
Saifeldeen Abdalmajeed Mahmood
Publication year - 2020
Language(s) - English
Resource type - Journals
ISSN - 1858-7313
DOI - 10.52981/fjes.v8i2.121
Subject(s) - distortion (music) , metric (unit) , artificial intelligence , pattern recognition (psychology) , computer science , similarity (geometry) , generalized normal distribution , image quality , gaussian , monotonic function , gaussian blur , image (mathematics) , mathematics , data mining , algorithm , image processing , statistics , normal distribution , image restoration , computer network , amplifier , mathematical analysis , operations management , physics , bandwidth (computing) , quantum mechanics , economics
When creating image quality assessment metric (IQA) no confirmation all distortion types are available. Non-specific distortion blind/no-reference (NR) IQA algorithms mostly need prior knowledge about anticipated distortions. This paper introduce a generic and distortion unaware (DU) approach for IQA with No Reference (NR). The approach uses three different measuring features which are initiated from the gist of natural scenes (NS) using Log-derivatives of the parameters; a general Gaussian distribution model, two sharpness functions, and Weibull distribution. All features were analyzed and co mpared together to examine their performance. When calibrating the proposed features performance on LIVE database, experiments show they have good contribution to the state of the art IQA and they outperform the popular full-reference peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) methods. Also they show sharpness features are the best when assess both prediction monotonicity and predict accuracy evaluation among the three features categories. Besides they show asymmetric generalized Gaussian distribution (AGGD) based features have the best correlation with differential mean opinion score.