NonLocalMeansFilter3d
Adaptive filter denoising a three-dimensional image.
Access to parameter description
For an introduction to image filters: Images Filtering.
This algorithm computes a weighted mean of a large set of voxels (search windows) around the voxel to be denoised (target voxel). The voxel weight in the search window is a function of the similarity between this voxel and the target voxel. A voxel very similar to the target voxel is assigned to a weight close to 1.0. A voxel very different from the target voxel is assigned to a weight close to 0 and has a small influence on the resulting denoised value.
The similarity between a voxel and the target voxel is based on the squared intensity differences between two patches of user-defined size, centered on the current voxel and the target voxel.
Images are often corrupted by noise because of the acquisition process. For an image $y$ defined on $\Omega$, which is the image domain, the noise added by acquisition is assimilated to an additive white gaussian noise: $$ y[k] = x[k] + e[k], \forall k \in \Omega $$ where $$ e[k] \sim \mathcal{N}(0, \sigma), \forall k \in \Omega $$
The denoised image $z$ is an estimation of $x$ as follows: $$ z[k] = \frac{\sum_{n\in\mathbb{Z}^3} w[k,n]\cdot y[k+n]}{\sum_{n\in\mathbb{Z}^3} w[k,n]} $$ with $$ w[k,n] = f(n)\exp\left(-\frac{ssd(k,n)}{\lambda}\right) $$ where:
The basic implementation of the previous equation has a complexity of $\mathcal{O}(P^3S^3NMQ)$ that can be reduced to $\mathcal{O}(S^3NMQ)$ where $N$, $M$ and $Q$ are, respectively, the X, Y, and Z dimensions of the input image, $P$ is the patch size, and $S$ is the search window size.
Reference:
A.Buades, B. Coll, J.M.Morel. "A Non-Local Algorithm for Image Denoising". Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), vol. 2, pp. 60-65, 2005.
See also
Access to parameter description
For an introduction to image filters: Images Filtering.
This algorithm computes a weighted mean of a large set of voxels (search windows) around the voxel to be denoised (target voxel). The voxel weight in the search window is a function of the similarity between this voxel and the target voxel. A voxel very similar to the target voxel is assigned to a weight close to 1.0. A voxel very different from the target voxel is assigned to a weight close to 0 and has a small influence on the resulting denoised value.
The similarity between a voxel and the target voxel is based on the squared intensity differences between two patches of user-defined size, centered on the current voxel and the target voxel.
Images are often corrupted by noise because of the acquisition process. For an image $y$ defined on $\Omega$, which is the image domain, the noise added by acquisition is assimilated to an additive white gaussian noise: $$ y[k] = x[k] + e[k], \forall k \in \Omega $$ where $$ e[k] \sim \mathcal{N}(0, \sigma), \forall k \in \Omega $$
The denoised image $z$ is an estimation of $x$ as follows: $$ z[k] = \frac{\sum_{n\in\mathbb{Z}^3} w[k,n]\cdot y[k+n]}{\sum_{n\in\mathbb{Z}^3} w[k,n]} $$ with $$ w[k,n] = f(n)\exp\left(-\frac{ssd(k,n)}{\lambda}\right) $$ where:
- $ssd(k,n)$ is the similarity criterion (sum of squared differences)
- $\lambda$ is the similarity
- $f(n)$ is the indicator function of the search region
- and $h(n)$ is the indicator function of the patch region.
The basic implementation of the previous equation has a complexity of $\mathcal{O}(P^3S^3NMQ)$ that can be reduced to $\mathcal{O}(S^3NMQ)$ where $N$, $M$ and $Q$ are, respectively, the X, Y, and Z dimensions of the input image, $P$ is the patch size, and $S$ is the search window size.
Reference:
A.Buades, B. Coll, J.M.Morel. "A Non-Local Algorithm for Image Denoising". Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR), vol. 2, pp. 60-65, 2005.
See also
Function Syntax
This function returns the outputImage output parameter.
// Function prototype.
std::shared_ptr< iolink::ImageView >
nonLocalMeansFilter3d( std::shared_ptr< iolink::ImageView > inputImage,
uint32_t patchRadius,
uint32_t kernelRadius,
double similarity,
NonLocalMeansFilter3d::KernelShape kernelShape,
std::shared_ptr< iolink::ImageView > outputImage = NULL );
This function returns the outputImage output parameter.
// Function prototype.
non_local_means_filter_3d( input_image,
patch_radius = 3,
kernel_radius = 6,
similarity = 0.6,
kernel_shape = NonLocalMeansFilter3d.KernelShape.CUBE,
output_image = None )
This function returns the outputImage output parameter.
// Function prototype.
public static IOLink.ImageView
NonLocalMeansFilter3d( IOLink.ImageView inputImage,
UInt32 patchRadius = 3,
UInt32 kernelRadius = 6,
double similarity = 0.6,
NonLocalMeansFilter3d.KernelShape kernelShape = ImageDev.NonLocalMeansFilter3d.KernelShape.CUBE,
IOLink.ImageView outputImage = null );
Class Syntax
Parameters
| Class Name | NonLocalMeansFilter3d |
|---|
| Parameter Name | Description | Type | Supported Values | Default Value | |||||
|---|---|---|---|---|---|---|---|---|---|
 |
inputImage |
The input image. | Image | Grayscale or Multispectral | nullptr | ||||
|
kernelShape |
The shape of the search window. The shape of the window search area can be used to improve the run time. For example, choosing a ball shape reduces the computation time by 50%.
|
Enumeration | CUBE | |||||
|
kernelRadius |
The search window half size side length or radius, in voxels.
In the case of a cube, a value N produces a cube of 2N+1 voxels side length. In the case of a ball, a value N produces a ball with a 2N+1 voxels diameter. The larger the search window, the better the results usually are. But the size of the search window also affects the run time significantly. This value has to be set to a large enough value so that similar structures can be found within the search window area. Values that are too small result in simple blurring of the image because there is not enough structural data within the search window area. |
UInt32 | >=1 | 6 | ||||
|
patchRadius |
The patch box half size side length, in voxels.
The weight of a voxel in the search windows is computed by comparing the neighborhood of this voxel with the neighborhood of the target voxel. This value represents the edge size of the neighborhood volume, in number of voxels (patch neighborhood is a cube), and affects the quality of the result. If this value is either much smaller, or much larger than the finest structures of the image, the algorithm shows few or no effect at all. This parameter has almost no effect on the computation time. |
UInt32 | Any value | 3 | ||||
|
similarity |
The similarity is used to compute weight w(k,n) assigned to each voxel in the search window.
The squared similarity is proportional to the standard deviation of the assumed gaussian noise of the input image. The larger this value, the more the resulting image is smoothed. The similarity has no effect on the computation time. |
Float64 | >=0.001 | 0.6 | ||||
 |
outputImage |
The output image. Its dimensions, type, and calibration are forced to the same values as the input. | Image | nullptr | |||||
Object Examples
auto foam = readVipImage( std::string( IMAGEDEVDATA_IMAGES_FOLDER ) + "foam.vip" ); NonLocalMeansFilter3d nonLocalMeansFilter3dAlgo; nonLocalMeansFilter3dAlgo.setInputImage( foam ); nonLocalMeansFilter3dAlgo.setPatchRadius( 3 ); nonLocalMeansFilter3dAlgo.setKernelRadius( 3 ); nonLocalMeansFilter3dAlgo.setSimilarity( 0.6 ); nonLocalMeansFilter3dAlgo.setKernelShape( NonLocalMeansFilter3d::KernelShape::CUBE ); nonLocalMeansFilter3dAlgo.execute(); std::cout << "outputImage:" << nonLocalMeansFilter3dAlgo.outputImage()->toString();
foam = imagedev.read_vip_image(imagedev_data.get_image_path("foam.vip"))
non_local_means_filter_3d_algo = imagedev.NonLocalMeansFilter3d()
non_local_means_filter_3d_algo.input_image = foam
non_local_means_filter_3d_algo.patch_radius = 3
non_local_means_filter_3d_algo.kernel_radius = 3
non_local_means_filter_3d_algo.similarity = 0.6
non_local_means_filter_3d_algo.kernel_shape = imagedev.NonLocalMeansFilter3d.CUBE
non_local_means_filter_3d_algo.execute()
print( "output_image:", str( non_local_means_filter_3d_algo.output_image ) )
ImageView foam = Data.ReadVipImage( @"Data/images/foam.vip" );
NonLocalMeansFilter3d nonLocalMeansFilter3dAlgo = new NonLocalMeansFilter3d
{
inputImage = foam,
patchRadius = 3,
kernelRadius = 3,
similarity = 0.6,
kernelShape = NonLocalMeansFilter3d.KernelShape.CUBE
};
nonLocalMeansFilter3dAlgo.Execute();
Console.WriteLine( "outputImage:" + nonLocalMeansFilter3dAlgo.outputImage.ToString() );
Function Examples
auto foam = readVipImage( std::string( IMAGEDEVDATA_IMAGES_FOLDER ) + "foam.vip" ); auto result = nonLocalMeansFilter3d( foam, 3, 3, 0.6, NonLocalMeansFilter3d::KernelShape::CUBE ); std::cout << "outputImage:" << result->toString();
foam = imagedev.read_vip_image(imagedev_data.get_image_path("foam.vip"))
result = imagedev.non_local_means_filter_3d( foam, 3, 3, 0.6, imagedev.NonLocalMeansFilter3d.CUBE )
print( "output_image:", str( result ) )
ImageView foam = Data.ReadVipImage( @"Data/images/foam.vip" ); IOLink.ImageView result = Processing.NonLocalMeansFilter3d( foam, 3, 3, 0.6, NonLocalMeansFilter3d.KernelShape.CUBE ); Console.WriteLine( "outputImage:" + result.ToString() );