Developing tools to process images for the Square Kilometre Array

[michael.h.f.wilkinson]

We have just had a paper published in IEEE Transactions on Image Processing, on a new method for processing images and volume data sets that the biggest (radio) telescopes like Meerkat and SKA are producing, or are going to produce. Many image processing tools lend themselves quite easily to parallel execution. However, the so-called connected morphological filters are very difficult in this respect. These latter have been found very useful at detecting faint structures in astronomical images. Previously we had managed to get good results for all kinds of images in shared-memory parallel machines, which works up to about 10 Gpixel or Gvoxel (I have a neat little 64 core 512 GB RAM machine at work for such chores), and on clusters for images up to 16 bits per pixel (maximum size processed to date: 165 Gpixel in under 5 minutes). The latest method can handle floating point images and volumes as well, albeit with some speed penalty. The full paper can be accessed freely here:

https://research.rug.nl/en/publications/distributed-connected-component-filtering-and-analysis-in-2-d-and

An earlier review on connected morphological filter for those interested can be found here:

http://www.cs.rug.nl/~michael/IEEE_SPM_2009_Salembier_Wilkinson.pdf