Structure similarity image quality assessment based on visual perception

Abstract

An image quality objective assessment method of wavelet domain structure similarity (WDSSIM) is proposed by considering human visual features for some deficiencies of spatial structural similarity (SSIM) assessment method while measuring the quality of geometric distorted images and noise polluted images. The method includes three steps: a) decomposing the reference image and the distorted image into sub-band images with different scales and frequencies by wavelet transform, and acquiring human visual in-band sensitivity and inter-band sensitivity according to human visual JND model; b) measuring the structural similarity between the same scale and frequency sub-band images of the reference image and distorted image respectively using in-band sensitivity as weights; c) obtaining structural similarity of the whole image by averaging the weighted SSIM of each pair subband images using inter-band sensitivity as weights. Experimental results show that WDSSIM method is agree with human visual features and its objective results are more consistent with the subjective assessment results than traditional methods.Keywords-human visual system (HVS); WDSSIM; human visual sensitivity; image quality objective assessment I.

Assessment of Different Image Transformation Methods on Diwata-1 Smi Images Using Structural Similarity Measure

“…This study attempted to find researches that present an itemized interpretation of SSIM indices to determine the most acceptable index value but was unsuccessful in this task. The researchers thus resorted to an indirect method by compiling several studies attempting to relate SSIM with subjective human visual perception (Lee & Lim, 2016;Wang, Wang, Liao & Lim, 2008;Lu, Bi & Wang, 2010;Yu & Liu, 2011) and obtaining the indices calculated from images with salt-andpepper noise, as this item is given low degradation mean opinion scores (DMOS) by human observers (Minamoto & Ohmura, 2014). From these studies, it was inferred that an SSIM rating of 0.8 can be considered as an acceptable similarity measure so that noisy images will fall below it.…”

Research on x-ray weld seam image quality evaluation methods

“…Currently, blackness determination relies on physical blackness meters, and there is a lack of a method to automatically determine blackness in X-ray weld image quality evaluation. Yu M [1] proposed a new machine learning-based full-reference image quality evaluation model, which includes four types of features: chromaticity, gradient, contrast sensitivity function, and Gaussian difference band, by fully simulating the human eye visual system and brain mechanism.Hu [2] incorporated the edge features and texture features of the image based on the weighted local entropy algorithm, and the experimental results showed that the method was consistent with subjective evaluation results with a high degree of consistency. Gao M [3] broke through the classical algorithmic framework based on local information and proposed a framework based on nonlocal information, and constructed a nonlocal gradient-based image quality evaluation algorithm within this framework.…”