Singal-indpendent and Signal-dependent noise modeling, parameter estimation and removal


Denoising is an intensively studied topic in the recent years. Improved as the denoising algorithms, they are not practical in use. In this research, we try to develop a robust, accurate and fast real noisy image denoising algorithm. Firstly, we studied the Additive White Gaussian Noise (AWGN) and then extended our reasearch to the practical signal-dependent noise.

Research projects

Noise level estimation for Additive White Gaussian Noise (AWGN)

AWGN noise model is widely used in the denoising literture, however, there are few research about the estimation of noise standard deviation.
In this research, we propose a fast and accurate algorithm to estimate the noise standard deviation from a single image.

Project page

Optimal noise level parameter for blind denoising

For denoising algorithms, even the noise standard deviation is perfectly estimated, they still cannot achieve the optimal PSNR or SSIM
due to the complex textures of a particular image. In this research, we propose an optimal noise level parameter to solve this problem.

Project page

Noise parameter estimation for signal-dependent noise

Noisy image from real camera sensor is better modeled as the generalized signal-dependent noise.
An accurate algorithm to estimate the three parameters of the signal-dependent nosie model is proposed in this research.

Project page

Signal-dependent noise (Real sensor noise) removal

A practical and high-quality signal-dependent noise (real sensor noise) algorithm is proposed in this research.
It is fully automatic, utilize the high-quality AWGN denoising algorithm and rely on only one signle noisy image.

Project page

Related publications