Colored Spatial Noise Overview

• Colored spatial noise is defined by its non-uniform power spectral density that produces long-range spatial correlations as opposed to uncorrelated white noise.
• It is mathematically constructed using power-law decay, convolution with spatial kernels, and spectral measures to ensure model regularity and well-posed stochastic equations.
• Applications include improving SPDE analysis, enhancing computational imaging techniques, and generating realistic textures in computer graphics.

Colored spatial noise refers to random fields defined over spatial domains whose covariance function exhibits nontrivial, typically long-range or structured correlations as opposed to spatially white (uncorrelated) noise. The prototypical mathematical models range from Gaussian and Lévy fields in stochastic partial differential equations (SPDEs) to non-Gaussian engineered ensembles in computational imaging and computer graphics. The spectral characteristics and associated correlation structure—collectively called “color”—play a central role in the regularity, physical realism, robustness, and computational efficiency of solutions and applications across mathematics, physics, and engineering.

1. Mathematical Construction of Colored Spatial Noise

Colored spatial noise arises by specifying a non-flat spectral power distribution for the underlying random field. For a zero-mean scalar field $N(x)$ on $\mathbb{R}^d$ or a discrete lattice, spatial color is characterized by its power spectral density (PSD) $S_N(k)$, which generally follows a power law: $S_N(k) \propto |k|^{-\alpha}$ where $k$ is the spatial frequency and $\alpha$ is the spectral exponent. Common examples are:

• $\alpha=0$: White noise (flat spectrum)
• $\alpha=1$: Pink noise (“$1/f$” noise), long-range positive correlations
• $\alpha<0$: Blue noise ($\mathbb{R}^d$0), high-frequency dominance and anti-correlations

A spatially colored Gaussian noise process $\mathbb{R}^d$1 can be defined as a centered Gaussian field with covariance

$\mathbb{R}^d$2

where $\mathbb{R}^d$3 is a symmetric, positive-definite kernel with spectral representation

$\mathbb{R}^d$4

and spectral measure $\mathbb{R}^d$5 for Riesz kernels, with $\mathbb{R}^d$6 controlling the singularity and decay rate (Chen et al., 2015, Kuzgun et al., 2022).

For Lévy colored noise, one convolves a Lévy white noise $\mathbb{R}^d$7 (constructed from a Poisson random measure of jumps with intensity $\mathbb{R}^d$8, $\mathbb{R}^d$9 a Lévy measure with finite variance) with a spatial kernel $S_N(k)$0: $S_N(k)$1 yielding an infinitely divisible process with spatial covariance determined by $S_N(k)$2 (Balan et al., 30 Apr 2025).

2. Spectral and Covariance Structure

The “color” of spatial noise is determined by the qualitative form of its covariance or spectral density:

• Riesz kernel: $S_N(k)$3, with spectral measure $S_N(k)$4
• Bessel/Matérn kernels: $S_N(k)$5, spectral decay controlled by $S_N(k)$6
• Heat kernel: smooth, rapidly-decaying correlations

Key admissibility is governed by the Dalang condition for SPDE stochastic integral well-posedness (Chen et al., 2015, Kuzgun et al., 2022): $S_N(k)$7 which restricts the singularity exponent (e.g., $S_N(k)$8 for Riesz kernels).

In applications such as computational imaging, colored speckle fields are synthesized in Fourier space via multiplication by $S_N(k)$9 with randomized phases, then inverse-transformed to real space (Nie et al., 2020, Nie et al., 2020, Li et al., 2021).

3. Stochastic Integration and Moment Estimates

The stochastic integral of predictable processes with respect to colored noise admits sharp moment estimates. For Lévy colored noise, Rosenthal's inequality provides $S_N(k) \propto |k|^{-\alpha}$0 bounds for such integrals (Balan et al., 30 Apr 2025): $S_N(k) \propto |k|^{-\alpha}$1 where $S_N(k) \propto |k|^{-\alpha}$2 (variance) and $S_N(k) \propto |k|^{-\alpha}$3 for higher moments.

For Gaussian colored noise, the isometry for stochastic integrals reduces to

$S_N(k) \propto |k|^{-\alpha}$4

where $S_N(k) \propto |k|^{-\alpha}$5.

Dalang’s condition guarantees the integrability needed for establishing existence, uniqueness, and regularity of random field solutions to SPDEs (Chen et al., 2015).

4. Applications in SPDE Theory

Colored spatial noise is central in the theory of SPDEs, dictating both regularity and qualitative phenomena such as intermittency, phase transition in Lyapunov exponents, and scaling laws of fluctuations:

• Well-posedness: Existence and uniqueness of solutions require the covariance kernel to satisfy Dalang’s condition (Chen et al., 2015, Balan et al., 30 Apr 2025).
• Spatial regularity: For driving noise with Riesz kernel $S_N(k) \propto |k|^{-\alpha}$6 in $S_N(k) \propto |k|^{-\alpha}$7, and fractional Laplacian generator $S_N(k) \propto |k|^{-\alpha}$8, the mild solution $S_N(k) \propto |k|^{-\alpha}$9 is Hölder continuous in space with exponent $k$0 provided $k$1 (Li, 2016).
• Moment growth and intermittency: For linear equations, the growth of moments and existence of phase transitions depend on the integrability properties of the spectral measure. For the parabolic Anderson model, explicit Lyapunov exponents and chaos expansions can be obtained (Balan et al., 30 Apr 2025).

In the case of Lévy colored noise, weak intermittency persists (i.e., growth of higher moments strictly exceeds that of second moments), and Poisson chaos expansions give explicit characterizations (Balan et al., 30 Apr 2025).

5. Engineered Colored Spatial Noise in Computational Imaging

Colored spatial noise is systematically engineered in computational imaging modalities—including computational ghost imaging (CGI), coded illumination, and super-resolving correlation imaging—by synthesizing speckle patterns with targeted spectral envelopes:

• Orthonormal colored-noise bases: Construction via Gram-Schmidt orthonormalization of colored speckle patterns yields statistically independent measurement modes. This approach, as in sub-Nyquist CGI, enables high-fidelity reconstructions at sampling ratios far below $k$2 compared to uncolored (white) noise patterns (Nie et al., 2020).
• Pink and blue noise speckles: In CGI, pink noise patterns (with long-range correlations) enhance robustness to ambient noise and pattern distortion, outperforming white noise in SNR under diverse interference (Nie et al., 2020). Blue noise, characterized by suppressed low-frequency content and negative nearest-neighbor intensity correlations, enables superresolving imaging: speckle fields with $k$3 achieve up to $k$4-fold improvement in transverse resolution over standard first-order imaging (Li et al., 2021).

Table: Spectral Exponents and Imaging Performance

Noise Type	Spectral Exponent $k$5	Properties / Application
White	$k$6	Uncorrelated; baseline for reference
Pink	$k$7	Robust under noise/interference (CGI)
Blue	$k$8, $k$9 in $\alpha$0	Anti-correlation; superresolution

6. Procedural and Learned Colored Spatial Noise in Graphics

In computer graphics, procedural and learned colored spatial noise models flexibly generate textures with spatially-varying color and complex spatial correlations. Recent advances using conditional denoising diffusion probabilistic models (DDPMs), with interpretable conditioning on user parameters and spatially-varying blending, allow generation and blending of multitype noise (e.g., Perlin, marble, Worley) as a unified framework (Maesumi et al., 2024). The network is conditioned via SPADE layers on spatial feature grids representing parameter blends, enabling one-pass generation of arbitrarily spatially-varying colored noise.

Applications include increasing the realism of textured assets, material design, and inverse procedural modeling, with significant improvements in perceptual fidelity compared to fixed-type procedural noise.

7. Physical and Mathematical Implications

In physical modeling, colored spatial noise arises naturally in hydrodynamic fluctuations (e.g., baryon diffusion with finite memory and spatial correlation length) (Kapusta et al., 2014). In such contexts, causal and analytic constraints on the spectral density of the noise—the so-called “colored” noise—guarantee both a finite propagation velocity and well-defined fluctuation-dissipation relations. In infinite-dimensional probability, the regularity, spatial roughness, and large-scale asymptotics of solutions to SPDEs are tightly linked to the precise choice of noise color (Huang et al., 2019, Li, 2016, Kuzgun et al., 2022).

Key parameter thresholds for the admissibility of colored spatial noise in SPDEs, including Dalang's condition (for Gaussian noise) and the corresponding moment and chaos estimates (for Lévy noise), provide the bridge from model construction to regularity and long-time analysis (Chen et al., 2015, Balan et al., 30 Apr 2025).

Colored spatial noise establishes a flexible and powerful framework for modeling spatially structured randomness at the core of stochastic analysis, physical modeling, imaging, and graphics. Its impact is governed by the interplay between the spectral decay, physical or application-specific constraints, and the analytic properties of the underlying equations or systems.