Skip to content

Filter-Based Denoising Module Guide

The Filter-Based Denoising module removes noise from TIFF image stacks using traditional mathematical filtering methods. It offers two approaches: Gaussian filtering for quick smoothing, and Non-Local Means (NLM) for edge-preserving denoising. No training is required—filters are applied directly to your images.

For a conceptual overview, see Filter-Based Denoising.

Filter Denoising module showing configuration optionsThe Filter Denoising module with method selection and parameter controls


Quick Start

  1. Launch module — Click "Filter-Based Denoising" from the workspace hub
  2. Select an image — Choose a TIFF stack from your workspace or use test data
  3. Choose a method — Select Gaussian (fast) or Non-Local Means (better quality)
  4. Adjust parameters — Configure filter strength and window sizes
  5. Process — Click "Start Denoising" and wait for completion
  6. View results — Open the denoised image in the Image Viewer

Step-by-Step Guide

Step 1: Data Selection

Select the image you want to denoise.

Step 1 showing file selectionStep 1 data selection with file chooser and test data option

Input Options

SourceDescriptionWhen to Use
Workspace FilesTIFF stacks uploaded to your workspaceYour own image data
Test DataBuilt-in noisy test imageLearning the module or testing settings

File Requirements

  • Format: Multi-page TIFF (.tif or .tiff)
  • Bit Depth: 8-bit or 16-bit grayscale
  • Channels: Single channel (grayscale)

Note: Single images are automatically treated as a 1-slice stack.

For more details, see Input Requirements.

Click "Next: Configure" to proceed to Step 2.


Step 2: Configure

Choose your denoising method and adjust its parameters.

Step 2 showing method selection and parametersStep 2 configuration with method selection and parameter controls

Method Selection

MethodSpeedEdge PreservationBest For
Gaussian FilterVery fastModerate (may blur edges)Quick denoising, uniform noise
Non-Local MeansSlowerExcellentFine details, structured images

For a detailed comparison, see Denoising Methods.


Gaussian Filter Parameters

The Gaussian filter smooths images by averaging neighboring pixels with weights that decrease with distance.

Gaussian parameter controlsGaussian filter controls for sigma and kernel size

ParameterRangeDefaultDescription
Sigma (σ)0.5 – 5.01.5Controls blur strength; higher = more smoothing
Kernel Size3, 5, 7, 9, 115Size of the filter window in pixels

Sigma Guidelines:

ValueEffect
0.5 – 1.0Subtle smoothing, preserves most detail
1.5 – 2.5Balanced noise reduction (recommended starting point)
3.0 – 5.0Strong smoothing, may blur edges

Kernel Size Guidelines:

SizeEffect
3×3Minimal smoothing, fastest
5×5Good balance (recommended)
7×7 – 11×11Stronger smoothing for high noise

For more details, see Gaussian Filter.


Non-Local Means Parameters

NLM finds similar patches throughout the image and averages them, preserving edges while reducing noise.

NLM parameter controlsNon-Local Means controls for filter strength and window sizes

ParameterRangeDefaultDescription
Filter Strength (h)1 – 3010Denoising intensity; higher = more noise removal
Template Window3, 5, 7, 9, 117Patch size for comparing similarity
Search Window11, 15, 21, 31, 4121Area to search for similar patches

Filter Strength Guidelines:

ValueEffect
1 – 5Subtle denoising, preserves all details
10 – 15Good balance for moderate noise (recommended)
20 – 30Aggressive denoising for very noisy images

Window Size Guidelines:

SettingSpeedQuality
Small windows (3, 11)FasterLess robust matching
Default (7, 21)BalancedGood for most images
Large windows (11, 41)SlowerMore robust for high noise

Tip: Start with default values and adjust based on results. NLM automatically adapts to your image's noise level.

For more details, see Non-Local Means.

Click "Next: Process" to proceed to Step 3.


Step 3: Process

Review your configuration and run the denoising process.

Step 3 showing configuration summary and start buttonStep 3 with configuration summary and Start Denoising button

Configuration Summary

Before processing, you'll see a summary of your settings:

  • Input File: Name of the selected image
  • Method: Gaussian or Non-Local Means
  • Parameters: All configured values

Processing

  1. Click "Start Denoising" to begin
  2. A spinner indicates processing is in progress
  3. Wait for completion (time depends on image size and method)

Processing Time Estimates:

MethodSmall Stack (50 slices)Large Stack (500 slices)
GaussianA few secondsUnder a minute
NLM1-2 minutes5-15 minutes

Note: NLM is significantly slower than Gaussian but produces better results for images with important edge details.


Results

After processing completes, you'll see the results summary.

Results section showing output details and action buttonsResults section with output details and action buttons

Results Display

InformationDescription
Output FileName of the denoised TIFF
Slices ProcessedNumber of slices in the stack
MethodWhich denoising method was used

Actions

ButtonAction
View in Image ViewerOpens the denoised result in the Image Viewer module
Process AnotherResets the module to denoise a different image

Output Files

FileDescriptionLocation
gaussian_denoised.tifGaussian-filtered outputresults/denoising/{processingId}/
nlm_denoised.tifNLM-filtered outputresults/denoising/{processingId}/

The output file:

  • Has the same dimensions as the input
  • Preserves the original bit depth (8-bit or 16-bit)
  • Is automatically added to your workspace for use in other modules

Choosing Between Methods

ConsiderationGaussianNon-Local Means
SpeedVery fastSlower
Edge preservationMay blur edgesExcellent
Fine detailsMay smooth overWell preserved
Noise typeUniform, random noiseAll noise types
Best forQuick preview, large stacksFinal results, important details

Recommendation: Start with Gaussian for a quick preview. If edges or fine details are important, use NLM for the final result.


Troubleshooting

IssueCauseSolution
Result still noisyParameters too conservativeIncrease sigma (Gaussian) or h (NLM)
Result too blurryParameters too aggressiveDecrease sigma or h; try NLM instead of Gaussian
Processing takes too longLarge stack with NLMUse Gaussian for preview; reduce search window for NLM
Edges are blurredUsing Gaussian on detailed imageSwitch to Non-Local Means method
"File not found" errorFile was moved or deletedRe-upload the file or select a different one

Module Overview:

Input:

Methods:

Part of DFG Priority Programme SPP2332 "Physics of Parasitism"