Method for reducing image or video noise

摘要

There is described a Minimal Iterativity Anisotropic Diffusion (MIAD) approach that estimates the required number of iterations N as a function of the image structure-under-noise ρ and the bound of the noise η. The time step λ is related to the image structure-under-noise ρ and to the bound of noise λ. The stopping time is calculated using T=λ·N, and the edge strength σ is determined as a function of T, η, and ρ.

主权项

1. A method for reducing noise in an image/video frame, comprising:
receiving an input image/video frame and a corresponding noise level η; determining a structure-under-noise estimator ρ as a function of a ratio between the noise level η and a structure measure parameter for the input image/video frame; determining a number of iterations N and a time step parameter λ from the structure-under-noise estimator ρ; determining a stopping time parameter T from the number of iterations N and the time step parameter λ; determining an edge strength parameter σ from the time step parameter λ, the noise level η, and the stopping time parameter T using:σ=1T·η+λmax⁢⁢ρ·1T·η.; denoising the input image/video frame by applying anisotropic diffusion thereto using the edge strength parameter σ, the time step parameter λ, and the number of iterations N, thereby obtaining a denoised image; and outputting the denoised image.