Information Technology Reference
In-Depth Information
image quality assessment. Furthermore, as disparity is an important attribute of
stereopsis, we will try to improve the performance of IQMs on stereoscopic image
quality assessment by integrating disparity information into the IQMs. We will
mainly focus on the full reference (FR) metrics in this study, which means that the
undistorted images are required for evaluating the quality of the distorted images.
3.1 Introduction to 2D Image Quality Metrics
Over the years, a number of researchers have contributed significant research in
the design of full reference image quality assessment algorithms, claiming to have
made headway in their respective domains [31]. In this study, eleven IQMs that
are summarized in Table 1 were employed and explained in detail as follows.
Table 1
Descriptions of image quality metrics
IQM
Descriptions
PSNR
Peak signal-to-noise ratio
SSIM
Single scale structural similarity
MSSIM
Multi-scale structural similarity
VSNR
Visual signal-to-noise ratio
VIF
Visual information fidelity
UQI
Universal quality index
IFC
Information fidelity criterion
NQM
Noise quality measure
WSNR
Weighted signal-to-noise ratio
PHVS
Modified PSNR based on HVS
JND
Just noticeable distortion model
PSNR is a traditionally used metric for visual quality assessment and still
widely used in evaluating the performance of compression and transmission
schemes. Although the performance of PSNR is worse than many other image
quality metrics in certain distortion types and respective domains, it is still appeal-
ing because it is simple to compute, has clear physical meanings, and is mathe-
matically convenient in the context of optimization.
SSIM (Structural SIMilarity) [15] is to compare structural information between
the reference and distorted images. Under an assumption that the human visual
system is highly adapted for extracting structural information from a scene, a simi-
larity measure can be constructed based on luminance comparison, contrast com-
parison, and structure comparison between the reference and distorted images.
MSSIM (Multi-scale SSIM) [32] is an extension of SSIM. MSSIM iteratively
applies a low-pass filter in the reference and distorted images and down-samples
the filtered images by a factor of 2. At each image scale
j
after
j
-1 iterations, the
contrast comparison and the structure comparison are calculated, respectively. The
