Multiscale structural similarity for image quality assessment

By Zhou Wang, Eero P. Simoncelli, A. Bovik. 2003

6,296 citations recorded by the source corpus.

Subfields: Computer Vision and Pattern Recognition, Media Technology, Evaluation & Benchmarks

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 Media Technology

In Evaluation & Benchmarks

Citations in the corpus

Evaluating Lossy Compression Rates of Deep Generative Models (2020, methodology): “However, for natural images, one could use more perceptually valid distortion metrics such as SSIM (Wang et al., 2004), MSSIM (Wang et al., 2003), or distances between deep features of a CNN (Johnson et al., 2016).”
The Unreasonable Effectiveness of Deep Features as a Perceptual Metric (2018, methodology): “This problem, often called full-reference image quality assessment (FR-IQA), has been a longstanding goal in the community, and there have been numerous perceptually motivated distance metrics proposed, such as SSIM [53], MSSIM [55], FSIM [57], and HDR-VDP [32].”
Learning deep representations by mutual information estimation and maximization (2018, methodology): “MS-SSIM correlated well with the MI estimate provided by MINE, indicating that these models encoded pixel-wise information well.”
Many Paths to Equilibrium: GANs Do Not Need to Decrease a Divergence At Every Step (2017, methodology): “We use the Inception Score [22] to measure how visually appealing CIFAR-10 samples are, the MS-SSIM metric [18, 24] to check sample diversity, and an Improved Wasserstein independent critic to assess overfitting, as well as sample quality [4].”
cGANs with Projection Discriminator (2018, methodology): “As we can see in Table 2, our projection model was able to achieve high inception accuracy and high MS-SSIM when compared to bicubic and concat .”
End-to-end Optimized Image Compression (2016, methodology): “Consistent with the appearance of these example images, we ﬁnd that distortion measured with a perceptual metric (MS-SSIM; Wang, Simoncelli, and Bovik, 2003), indicates substantial improvements across all tested images and bit rates (ﬁgure 7; additional examples provided in the appendix and online).”
CERTIFAI: Counterfactual Explanations for Robustness, Transparency, Interpretability, and Fairness of Artificial Intelligence models (2019, methodology): “Hence, we use SSIM (Structural Similarity Index Measure) [Wang et al., 2003], which has been shown to be a better measure of what humans consider to be similar images [Wang et al., 2002].”
Recent Progress on Generative Adversarial Networks (GANs): A Survey (2019, methodology): “Different from single scale SSIM [36] metric, which is used to measure the similarity between two images, Wang et al. [37] ﬁrstly proposed the MS-SSIM for multiple scales image quality assessment.”
High-Fidelity Generative Image Compression (2020, background): “, (Multi-Scale) Structural Similarity Index ((MS-)SSIM) [53, 52], Learned Perceptual Image Patch Similarity (LPIPS) [57]), the weakness of each metric is exploited by the learning algorithm, e.”
Soft-to-Hard Vector Quantization for End-to-End Learning Compressible Representations (2017, methodology): “SSIM and MS-SSIM are the structural similarity index [37] and its multi-scale SSIM computed variant [36] proposed to measure the similarity of two images.”
Full Resolution Image Compression with Recurrent Neural Networks (2016, methodology): “We use Multi-Scale Structural Similarity (MS-SSIM) [19], a well-established metric for comparing lossy image compression algorithms, and the more recent Peak Signal to Noise Ratio - Human Visual System (PSNR-HVS) [7].”
The Perception-Distortion Tradeoff (2017, methodology): “Many alternative, more perceptual, distortion measures have been proposed, including SSIM [40], MS-SSIM [42], IFC [36], VIF [35], VSNR [3] and FSIM [47].”
Learned Video Compression (2018, methodology): “On the top row of Figure 8, we present the average MS-SSIM across all videos for each dataset and for each codec (Section 5.2), as function of BPP. Relative compressed sizes.”
Image Super-Resolution Using Deep Convolutional Networks (2014, methodology): “Following [42], super-resolution is only applied on the luminance channel (Y channel in YCbCr color space) in Sections 4.1- 4.4, so c = 1 in the first/last layer, and performance (e.g., PSNR and SSIM) is evaluated on the Y channel.”
The GAN Landscape: Losses, Architectures, Regularization, and Normalization (2018, methodology): “[30] Zhou Wang, Eero P Simoncelli, and Alan C Bovik.”
Lossy Image Compression with Compressive Autoencoders (2017, methodology): “We used the implementation of van der Walt et al. (2014) for SSIM and the implementation of Toderici et al. (2016b) for MS-SSIM.”
Variational Approaches for Auto-Encoding Generative Adversarial Networks (2017, methodology): “[29] assess the similarity between class-conditional generated samples using MSSSIM [40], an image similarity metric that has been shown to correlate well with human judgement.”