The statistics of natural images

By Daniel L. Ruderman. 1994

How researchers see this paper

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

Citations in the corpus

Density Modeling of Images using a Generalized Normalization Transformation (2015, background): “Variants of divisive normalization have been found to reduce dependencies when applied to natural images or sounds and to produce approximately Gaussian responses (Ruderman, 1994; Schwartz & Simoncelli, 2001; Malo & Laparra, 2010).”
In search of illumination invariants (2001, methodology): “15 [40, 41], lends empirical support for the technique, letting the albedo of the object approach zero, employed in the proof of Theorem 3.”
On Advances in Statistical Modeling of Natural Images (2004, methodology): “This non-Gaussian behavior of images has also been studied and modeled by Ruderman [69], Simoncelli and Adelson [76], Moulin and Liu [60], and Wainwright and Simoncelli [88].”
Wavelet-based image estimation: an empirical Bayes approach using Jeffrey's noninformative prior (2001, background): “Other types of invariance, namely spatial-scale invariance (or self similarity), however, have received considerable attention (see Field [14] and Ruderman [33]).”
SAN: Learning Relationship between Convolutional Features for Multi-Scale Object Detection (2018, background): “Several researchers studied natural image statistics and the relationship between two resampled images [11,30,29].”
Learning to Maintain Natural Image Statistics (2018, background): “To highlight the difference between the groups, in their ability to generate natural looking images, we revisit a decades-old observation, which says that natural images exhibit a typical corresponding power-spectrum [1,41].”
A Feature-Enriched Completely Blind Image Quality Evaluator (2015, background): “The so-called MSCN coefficients I (i, j ) have been observed to follow a unit normal distribution on natural images that have not suffered noticeable quality distortions [29] .”
Analyzing Neural Responses to Natural Signals: Maximally Informative Dimensions (2002, background): “7(a), which are qualitatively similar to those of light intensities in natural scenes (Voss & Clarke, 1975; Ruderman, 1994).”
Natural image statistics and efficient coding. (1996, unknown): “The approach that we and others [1, 7, 17 ] have recently taken is to look at the problem from the opposite end, and to study the structure of the images we typically view.”
Saliency Detection: A Spectral Residual Approach (2007, background): “Log spectrum representation Of the invariant factors of natural image statistics, scale invariance is the most famous and most widely studied property [20, 17].”
No-reference quality assessment using natural scene statistics: JPEG2000 (2005, background): “…computer generated graphics scenes, cartoons, and animations, paintings and drawings, random noise, or images and videos captured from nonvisual stimuli such as radar and sonar, X-rays, ultrasounds, etc. Natural scenes form an extremely tiny subset of the set of all possible scenes [21], [22].”
Occlusion Models for Natural Images: A Statistical Study of a Scale-Invariant Dead Leaves Model (2004, methodology): “Below, we investigate the second-order statistics and scaling of an occlusion model with ﬁnite cut-offs.”
Face recognition: A literature survey (2003, background): “It is well known that there exist signi.cant statistical redundancies in natural images [Ruderman 1994].”
Image denoising using scale mixtures of Gaussians in the wavelet domain (2003, background): “Contemporary models of image statistics are rooted in the television engineering of the 1950s (see [4] for review), which relied on a characterization of the autocovariance function for purposes of optimal signal representation and transmission.”
Blind/Referenceless Image Spatial Quality Evaluator (2011, methodology): “Ruderman observed that local mean subtraction and variance normalization on log-contrast values [11] produces decorrelated coefficients which follow gaussian like distributions.”
Fully Deep Blind Image Quality Predictor (2017, background): “Ruderman observed that locally normalized luminance values tend towards a unit normal Gaussian characteristic for natural images [48].”
The `independent components''of natural scenes are edge filters (1997, unknown): “As empasized by Ruderman (1994) and Field (1994), the general form of these histograms is double-exponential (e ?ju i j ), or`sparse', meaning peaky with a long tail, when compared to a gaussian .”
No-Reference Image Quality Assessment in the Spatial Domain (2012, methodology): “The regularity of natural scene statistics (NSS) has been well established in the visual science literature, where regularity has been demonstrated in the spatial domain [15], and in the wavelet domain [16].”