Why You Should Care About A Compressive Sensing Approach to CT, Emil Sidky and his co-authors Jakob H. Jørgensen, Per Christian Hansen, Ingrid S. Reiser, and Xiaochuan Pan looked at a model for breast CT. I then asked Emil if somehow a similar phantom could be released to the world so that it gets some sort of trial by any of the algorithms being developed for CS or not. Emil kindly responded with:
The theory of how to get this type of phantom is in this paper by Reiser and Nishikawa:
Although I did make a few unimportant modifications for this particular image. I excluded the small micro-calc objects because they are there to test discretization error, so the data for those structures are generated from a continuous object model. If somebody follows up on this CS basis problem, we can discuss that and other sources of error typical to CT later. I scaled the image to [0,65535] in order to export it in unsigned 2-byte format.To recover it:(1) gunzip it(2) read in the 1024x1024 and convert to floats(3) multiply by 1.15/65535.0By the way, since we're talking about CS, I noticed that gzip gets the image down to 42K.What about a gzip basis ? :)
The file is here. Thanks Emil for following up on this.