Wang preservation. Research and rapid mixture of experts model image super-resolution method based on [D]. 2015.
PSNR and SSIM in sometimes not very precise image representation quality standards, according to the paper in a certain amount of human perception score as a reference, using Spearman's rank correlation coefficient to verify the validity of each image quality evaluation, the goal is find more in line with standard image quality evaluation of the human eye. In addition to PSNR and SSIM, the paper also compares several other image quality evaluation criteria:
Fidelity information (Information Fidelity Criterion, IFC) [43],
Multiscale structural similarity (Multi-scale Structure Similarity Index, MSSIM) [44],
Noise Quality Evaluation (Noise at Quality the Measure, NQM ) [45],
Global Quality Evaluation (Universal Quality Index, UQI) [46],
Weighted PSNR (Peak Signal-to-Weighted Noise Ratio, WPSNR ),
Fidelity visual information (Visual Information Fidelity, VIF) [47].
Wherein, the IFC statistical indicators by natural multiscale image with a Gaussian mixture model (Gaussian Scale Mixtures) obtained information is used to measure the degree of damage, the main feature is a waveform for the high frequency information, instead of the human eye is not sensitive low-frequency information. The study showed, IFC index and the human eye perceives the results of the closest, followed by NQM indicators.
In the super-resolution reconstruction, in order to better reflect the reconstructed image quality is good or bad, in addition to the commonly used image restoration quality standard PSNR and SSIM but also the application of IFC, NQM (Noise Quality Measure), WPSNR (Weighted Peak Signal- to-noise Ratio), MSSSIM (Multi-Scale Structure Similarity Index) and SAM (Spectral Angle Match) evaluation criteria, which are commonly used for image restoration information is good or bad quality of the evaluation criteria. The higher their value, the higher in general, representative of the quality of image restoration.
Road passenger Baba - Hyperspectral image information convolution neural network technology based recovery http://www.doc88.com/p-7768661377035.html
SRCNN in:
Test set. Set5 (5 Zhang images), Set14 (14 Images) and BSD200 (200 Images) was used to assess the performance of upscaling is the 2,3,4.
Evaluation Matrix. In addition to PSNR and SSIM indicators are widely used, we also use a additional four evaluation matrix, called the IFC, NQM quality measurement noise, the signal to noise ratio of the weight-point similarity score MSSSIM WPSNR multiscale structure, as in [44] It said, it human perception scores are highly correlated.
【other】
Overview of image quality assessment (assessment indicators, traditional detection methods) - qq_23304241's blog - CSDN blog https://blog.csdn.net/qq_23304241/article/details/80953613
NQM (Noise Quality Measure)
Image Quality Assessment Based on a Degradation Model http://users.ece.utexas.edu/~bevans/papers/2000/imageQuality/index.html