Optimized-Bayesian-Nonlocal-means-with-block(OBNLM)

Optimized bayesian nonlocal-means algorithm for denoising ultrasound image

Description

• BayesianNLM.m : the OBNLM algorithm
• getPearsonDistance.m: compute the pearson distance based bayesian framework
• ImgNormalize.m: pre-processing the image (histogram stretching)
• testBayesianNLM.m: test our algorithm using the input (noisyImage.png) and get the output (despeckledImage.png)

Reference Paper

Coupé, Pierrick, et al. "Nonlocal means-based speckle filtering for ultrasound images." IEEE transactions on image processing 18.10 (2009): 2221-2229.

Basic Principle

The blockwise Nonlocal Means algorithm is completed. The basic principle is shown below:

Object Function: <img src="http://latex.codecogs.com/svg.latex?NL(u)(B_j) = \sum_{i\in\Delta_j}w(B_i,B_j)u(Bi)" border="0"/> with <img src="http://latex.codecogs.com/svg.latex? w(B_i,B_j)=\frac{1}{Z^j}e^{-\frac{dp(u(B_i),u(B_j))}{h^2}}" border="0"/>

<p align="center"><img src="http://latex.codecogs.com/svg.latex? B" border="0"></p>: the blockwise of image

<img src="http://latex.codecogs.com/svg.latex? dp(u(B_i),u(B_j))" border="0"/>:the pearson distance between blockwise

Results

The original ultrasound image (with speckle) vs the despeckled ultrasound image:

<p align="center"> <img src="noisyImage.png" border="0" height = "380" width="400"/> <img src="despeckledImage.png" border="0" height="380" width="400"/> </p>