Local Approximation Signal and Image Processing

Spatially Adaptive Non-Gaussian Imaging Via Fitted Local Likelihood Technique

for MATLAB version 6.5 or later

The MATLAB routine Fitted Local Likelihood (FLL) implements a new technique for spatially adaptive filtering. This statistics is proposed for selection of an adaptive size estimation neighborhood and based on the local maximum likelihood estimation. The algorithm is developed for quite general observation models subject to the class of the exponential distributions. This algorithm shows a better performance than the intersection of confidence interval (ICI) algorithm, in particular, for Poissonian data.

In many applications the noise that corrupts the signal is non-Gaussian and signal-dependent. There is a variaty of heuristic adaptive-neighborhood approaches for filtering signal and images corrupted by signal-dependent noise. The procedure is given for observations subject to the class of exponential distributions which includes the Poissonian model as an important special case.

The performance of the algorithm is illustrated for image denoising with data having Poissonian, Gaussian and Bernoulli observations.

The main algorithm is prepared as demo, so that it can be executed in a straightforward manner. This demo reproduces figures and results from the paper:
PDF Katkovnik V., and V. Spokoiny, “Spatially Adaptive Non-Gaussian Imaging via Fitted Local Likelihood Technique.”, Proc. Int. TICSP Workshop Spectral Meth. Multirate Signal Process., SMMSP 2006, Florence, 2006.

The provided demo is open-source, and may be modified and tuned to be exploited with other data. This DemoBox is available free-of-charge for educational and non-profit scientific research, enabling others researchers to understand and reproduce our work. Any unauthorized use of the LASIP routines for industrial or profit-oriented activities is expressively prohibited.

The main routine provided in this DemoBox is the following:


This demo performs denoising based on the novel statistical multiple hypothesis testing rule. The data observed is considered to be corrupted with Gaussian and non-Gaussian (signal-dependent) types of noise. The Poissonian model of noise is of great importance.
The technique incorporates the signal-adaptive FLL denoising.
This software is based (and requires) the LASIP image restoration demobox.
Any unauthorized use of the LASIP routines for industrial or profit-oriented activities is expressively prohibited. By downloading any of the LASIP files, you implicitly agree to all the terms of the LASIP limited license PDF.


Tampere University of Technology - Department of Signal Processing - Transforms and Spectral Methods Group