Fast 2D/3D morphological analysis on GPU
A library for fast 2D/3D volume analysis using CUDA — from simple but fast morphological operations and segmentation to efficient watershed, local thickness and fiber analysis
Customer:
Neoscan
(Belgium)
Platform:
Windows, Linux

Technology stack:
C++ 17, CUDA
Software description

A set of highly optimized CUDA routines for fast morphological analysis of large 2D images and 3D volumes.

Software supports large datasets exceeding available GPU memory.

While originally developed for Windows, the Linux version offers about 3 times better performance because of better CUDA memory management on this operating system.
Key features

  • Fast large 2D image and 3D volumes processing with CUDA
  • Binarization and thresholding
  • Classical erode/dilate of binary images by a spherical kernel
  • High performant erode/dilate algorithm independent of kernel radius
  • Binary image segmentation with no additional 3D volume memory usage
  • Particle characterization and statistics
  • Operations on separate particles (pore filling, removing small particles, etc.)
  • Watershed separation using water-drop approach
  • Distance field calculation
  • Local thickness estimation
  • Determination of preferred fiber orientation
  • Volume-of-interest and masking support