ITK/Summer v4 2011 Meeting/A2D2/Deconvolution

How our A2D2 will advance the field (of microscopy)

• Deconvolution is commonly used to reduce blur in microscopy images
• We provide algorithms to do this and data to evaluate them

Distribution

• Code
  • New Deconvolution module
  • Point-spread function models for microscopy - external module
• Data: MIDAS

• Dependencies
  • FFT-based convolution filter (Cory will do, modification of Gaetan's Insight Journal contribution)

New Features and Classes in ITK

Two widefield microscope point-spread function models

• itk::MicroscopePointSpreadFunctionImageSource.h
  • itk::OPDBasedWidefieldMicroscopePointSpreadFunctionImageSource
    • itk::GibsonLanniPointSpreadFunctionImageSource
    • itk::HaeberlePointSpreadFunctionImageSource
• itk::OPDBasedWidefieldMicroscopePointSpreadFunctionIntegrand
• itk::BeadSpreadFunctionImageSource
• itk::ScanImageFilter
• itk::SphereConvolutionFilter

Classes to enable fitting of parametric image models via the registration framework

• itk::ParametricImageSource
• itk::ImageToParametricImageSourceMetric

Four deconvolution algorithms

• • Wiener filter
  • Landweber algorithm
  • Richardson-Lucy maximum likelihood estimation
  • Parametric semi-blind deconvolution using parametric point-spread function model

New Features and Classes in ITK

Two widefield microscope point-spread function models

• itk::MicroscopePointSpreadFunctionImageSource.h
  • itk::OPDBasedWidefieldMicroscopePointSpreadFunctionImageSource
    • itk::GibsonLanniPointSpreadFunctionImageSource
    • itk::HaeberlePointSpreadFunctionImageSource
• itk::OPDBasedWidefieldMicroscopePointSpreadFunctionIntegrand
• itk::BeadSpreadFunctionImageSource
• itk::ScanImageFilter
• itk::SphereConvolutionFilter

Classes to enable fitting of parametric image models via the registration framework

• itk::ParametricImageSource
• itk::ImageToParametricImageSourceMetric

Four deconvolution algorithms

• • Wiener filter
  • Landweber algorithm
  • Richardson-Lucy maximum likelihood estimation
  • Parametric semi-blind deconvolution using parametric point-spread function model

Data

Estimated size

1 GB

Sharing

• Hosting on Kitware MIDAS installation
• Some images contributed as test data

Images

Synthetic images of simple objects (50 images)

• Spheres and cylinders convolved with example PSF
• Corrupted by various levels of noise
• 2 shapes X 5 different sizes for each shape X 5 different signal-to-noise ratios

Measured images of spherical beads (15 images )

• We have widefield images of 100nm, 200nm, and 500nm beads from Dr. Kerry Bloom (size verified with scanning electron microscope)
• 5 images of each bead size

Collaborator images (15-30 images)

• 5 images and their associated PSFs from each collaborator (Bloom, Weinbert, and Superfine.)