Scope of validity of PSNR in image/video quality assessment

By Q. Huynh-Thu, Mohammed Ghanbari. 2008

3,042 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Signal Processing

View Paper: https://www.semanticscholar.org/paper/dad5932e2a40c87d17e40cf8d072ffdb43046fd9

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Signal Processing

Citations in the corpus

All in One Bad Weather Removal Using Architectural Search (2020, methodology): “Although our method does not produce the best result in terms of the PSNR, we still achieve a competitive performance without incorporating extra information such as that obtained from the edge attention mechanism in [42].”
A Model-Driven Deep Neural Network for Single Image Rain Removal (2020, methodology): “Note that as the human visual system is sensitive to the Y channel of a color image in YCbCr space, we compute PSNR and SSIM based on this luminance channel.”
Hyperspectral Image Denoising Employing a Spatial–Spectral Deep Residual Convolutional Neural Network (2018, methodology): “21 reveals the quantitative evaluation results (MPSNR) with different numbers of adjacent spectra K (the horizontal axis represents a half value of K ) in the simulated experiment.”
Deep Joint Rain Detection and Removal from a Single Image (2016, methodology): “For the experiments on synthesized data, two metrics Peak Signal-to-Noise Ratio (PSNR) [19] and Structure Similarity Index (SSIM) [33] are used as comparison criteria.”
Peak signal-to-noise ratio revisited: Is simple beautiful? (2012, background): “It has been reported that PSNR estimates the relative perceptual quality reasonably well when fixed content and codec is concerned across different test cases [3].”
Image De-Raining Transformer (2022, methodology): “Following previous de-raining methods, peak-signal-to-noise ratio (PSNR) [103] and structure similarity (SSIM) [98] are employed for evaluating model performance.”
Efficient Parallel Framework for HEVC Motion Estimation on Many-Core Processors (2014, methodology): “Although there are other video quality assessment approaches [40], the PSNR-based video quality measurement is widely used during the development of HEVC [41].”
Image De-raining via Continual Learning (2021, methodology): “Moreover, peak-signal-to-noise ratio (PSNR) [17] and structure similarity (SSIM) [46] are employed for evaluating the model performance.”
Invisible Backdoor Attack with Sample-Specific Triggers (2020, methodology): “Besides, we adopt the peak-signal-tonoise-ratio (PSNR) [14] and `∞ norm [12] to evaluate the stealthiness.”
Single Image Deraining using Scale-Aware Multi-Stage Recurrent Network (2017, methodology): “To quantitatively evaluate these methods, four evaluation metrics are used: Peak Signal-to-Noise Ratio (PSNR) [10], Structure Similarity Index (SSIM) [22], Visual Information Fidelity (VIF) [19], and Feature Similarity Index (FSIM) [24].”
Transframer: Arbitrary Frame Prediction with Generative Models (2022, methodology): “Metrics For video generation we use the following metrics: Structural Similarity Index Measure (SSIM, [60]), Peak Signal-to-noise Ratio (PSNR, [27]),”
COVERAGE — A novel database for copy-move forgery detection (2016, methodology): “Amongst existing metrics, we consider Peak Signal-to-Noise Ratio (PSNR) [12] (in decibels) and structural similarity (SSIM) [13] computed on 3 × 3 overlapping windows of the forged and original images, as global similarity metrics.”
Learning A Sparse Transformer Network for Effective Image Deraining (2023, methodology): “By adding local feature extraction and fusion at different scales, the MSFN can indeed better boost the performance, and achieve PSNR gain of 0.21 dB over GDFN. Effectiveness of MEFC .”
All-In-One Image Restoration for Unknown Corruption (2022, background): “For example, AirNet is 1.4 and 0.0074 higher than the best method under the OBO setting in PSNR and SSIM, respectively.”
Shift-Net: Image Inpainting via Deep Feature Rearrangement (2018, background): “s of SSIM and mean ‘ 2 loss, our Shift-Net is a little inferior to MNPS [47] but are better than the others. Actually, both PSNR and SSIM are correlated poorly with human assessment of visual quality [12,14,24,36,42]. Moreover, inpainting aims at generating visual-pleasing results, rather than faith”
Joint Rain Detection and Removal from a Single Image with Contextualized Deep Networks (2020, methodology): “We look further into the beneﬁt of the contextualized dilated , LP [21], DetailNet [1], DID-MDN [66], ID-CGAN [67], UGSM [64], Proposed.”
Stochastic Variational Video Prediction (2017, background): “In both our qualitative and quantitative comparisons, SV2P produces substantially improved video predictions when compared to the same model without stochasticity, with respect to standard metrics such as PSNR and SSIM.”