Internal statistics of a single natural image

By Maria Zontak, Michal Irani. Computer Vision and Pattern Recognition, 2011

390 citations recorded by the source corpus.

Subfields: Computational Mechanics, Computer Vision and Pattern Recognition

View Paper: https://www.semanticscholar.org/paper/68c74264cf133638956f43fd356dae5a512996b5

How researchers see this paper

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

In Computational Mechanics

In Computer Vision and Pattern Recognition

Citations in the corpus

Zero-Shot Super-Resolution Using Deep Internal Learning (2017, background): “In fact, it was empirically shown by [24] that the internal entropy of patches inside a single image is much smaller than the external entropy of patches in a general collection of natural images.”
Noise2Void - Learning Denoising From Single Noisy Images (2018, background): “Instead, they can be directly applied to a test image where they extract all required information [27].”
Light Field Super-Resolution with Zero-Shot Learning (2021, background): “However, as analyzed in [53], in a natural image, the more gradient contents are within a patch, the less this patch recurs across scales.”
Patch Based Synthesis for Single Depth Image Super-Resolution (2012, background): “In the single image SR community, quantitative results have been criticized for not being representative of perceptual quality [17,18], and some authors choose to ignore them completely [19,21,20].”
Blind Deblurring Using Internal Patch Recurrence (2014, background): “It was empirically shown by [7,20] that almost any small image patch in a natural image (5× 5 or 7 × 7) re-appears “as is” (without shrinking the patch) in smaller scaled-down versions of the image (Fig.”
Training-Free, Single-Image Super-Resolution Using a Dynamic Convolutional Network (2018, background): “The assumption is justified given the interscale correlation and recurrence of image patches [33], [34].”
Crisp Boundary Detection Using Pointwise Mutual Information (2014, positive): “The utility of internal image statistics has been previously demonstrated in the context of super-resolution and denoising [29] as well as saliency prediction [30].”
Neural Nearest Neighbors Networks (2018, background): “They rely on similar structures being more likely to encounter within an image than across images [53].”
Single-Image Super-Resolution: A Benchmark (2014, methodology): “The priors of cross-scale self-similarity used in the GBI method [11] is also a specialized statistical prior as shown in [9,49].”
Learning to Maintain Natural Image Statistics (2018, background): “This has been shown effective for a range of applications [17,18,19,20].”
Single image super-resolution from transformed self-exemplars (2015, background): “Internal dictionaries have been shown to contain more relevant training patches, as compared to external dictionaries [44].”
Maintaining Natural Image Statistics with the Contextual Loss (2018, background): “nt approach is to prefer image-specic priors, learned only from the image to be restored, without any external information from other images. This has been shown eective for a range of applications [17,18,19,20]. Recently, attempts have been made to utilize image-specic priors within the CNN framewo”
SinGAN: Learning a Generative Model From a Single Natural Image (2019, background): “Modeling the internal distribution of patches within a single natural image has been long recognized as a powerful prior in many computer vision tasks [64].”
Nonparametric Blind Super-resolution (2013, methodology): “Our method is based on the universal property that natural image patches tend to recur abundantly, both across scales of the same image, as well as in an external database of other natural images [22, 8, 7].”
Deep Learning for Single Image Super-Resolution: A Brief Review (2018, background): “As demonstrated in [101], the visual entropy inside a single image is much smaller than the large training dataset collected from wide ranges, so unlike external-example SISR algorithms, a very small CNN is sufficient.”
Patch Complexity, Finite Pixel Correlations and Optimal Denoising (2012, background): “The explanation of [22] is that for textured areas the probability of finding relevant neighbors within the local neighborhood patches is too low.”
Contrastive Learning for Unpaired Image-to-Image Translation (2020, methodology): “Single image internal statistics has been proven effective in many vision tasks such as segmentation [30], super-resolution, and denoising [84,61].”