Product of Gaussian Mixture Diffusion Models

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zach, Martin, Kobler, Erich, Chambolle, Antonin, Pock, Thomas
Format: Preprint
Published: 2023
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914636382601216
author Zach, Martin
Kobler, Erich
Chambolle, Antonin
Pock, Thomas
author_facet Zach, Martin
Kobler, Erich
Chambolle, Antonin
Pock, Thomas
contents In this work we tackle the problem of estimating the density $ f_X $ of a random variable $ X $ by successive smoothing, such that the smoothed random variable $ Y $ fulfills the diffusion partial differential equation $ (\partial_t - Δ_1)f_Y(\,\cdot\,, t) = 0 $ with initial condition $ f_Y(\,\cdot\,, 0) = f_X $. We propose a product-of-experts-type model utilizing Gaussian mixture experts and study configurations that admit an analytic expression for $ f_Y (\,\cdot\,, t) $. In particular, with a focus on image processing, we derive conditions for models acting on filter-, wavelet-, and shearlet responses. Our construction naturally allows the model to be trained simultaneously over the entire diffusion horizon using empirical Bayes. We show numerical results for image denoising where our models are competitive while being tractable, interpretable, and having only a small number of learnable parameters. As a byproduct, our models can be used for reliable noise level estimation, allowing blind denoising of images corrupted by heteroscedastic noise.
format Preprint
id arxiv_https___arxiv_org_abs_2310_12653
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Product of Gaussian Mixture Diffusion Models
Zach, Martin
Kobler, Erich
Chambolle, Antonin
Pock, Thomas
Image and Video Processing
In this work we tackle the problem of estimating the density $ f_X $ of a random variable $ X $ by successive smoothing, such that the smoothed random variable $ Y $ fulfills the diffusion partial differential equation $ (\partial_t - Δ_1)f_Y(\,\cdot\,, t) = 0 $ with initial condition $ f_Y(\,\cdot\,, 0) = f_X $. We propose a product-of-experts-type model utilizing Gaussian mixture experts and study configurations that admit an analytic expression for $ f_Y (\,\cdot\,, t) $. In particular, with a focus on image processing, we derive conditions for models acting on filter-, wavelet-, and shearlet responses. Our construction naturally allows the model to be trained simultaneously over the entire diffusion horizon using empirical Bayes. We show numerical results for image denoising where our models are competitive while being tractable, interpretable, and having only a small number of learnable parameters. As a byproduct, our models can be used for reliable noise level estimation, allowing blind denoising of images corrupted by heteroscedastic noise.
title Product of Gaussian Mixture Diffusion Models
topic Image and Video Processing
url https://arxiv.org/abs/2310.12653