rjw57
Make library useful to others
@scf32 notes that the some tasks need to be done for dtcwt to be useful to him. I have reproduced them below:
- [ ] Double sampling rate of transform in all 3 dimensions (scale dimension can be handled fairly trivially at a higher level by initial scaling of the input image and interleaving the pyramid outputs, but the spatial dimensions require a modified colfilter).
- [ ] Ability to work in RGB colour space. All I've done to achieve this is merge the low-pass outputs from the 3 colour channels at level 1 and assume the colour distances in the RGB cube propagate through the rest of the transform.
- [ ] Subsample keypoint refinement using a 4x4 window method. This is not related to the DTCWT itself but just FYI.
@scf32 Re RGB colour space: I think it is correct to simply transform each colour plane. I suspect the nicest behaviour would be for the 2D transform to accept 3D arrays and simply transform each slice independently. Opinions?
@scf32 Can you point me at some MATLAB code which does the double sampling rate stuff?
@scf32 I'm unfamiliar with the 4x4 refinement. TBH, the existing keypoint stuff in dtcwt id a little bit experimental and probably needs a re-write.
Double sampling low level functions are in Tim's git.csx.
The 4x4 refinement is one of Pashmina's creations. The current fastest version of that should also be in Tim's repo.
Yes. Actually, for reference it's best to look at refineMaxima(), which is the original we're working from. If that's not there then I can send a copy your way.
On 8 August 2014 13:38, Rich Wareham [email protected] wrote:
Tim's repo being this https://git.csx.cam.ac.uk/i/eng-sigproc/u/teg28/MatlabDTCWT.git private one, yes?
— Reply to this email directly or view it on GitHub https://github.com/rjw57/dtcwt/issues/105#issuecomment-51595121.
On 8 August 2014 13:57, scf32 [email protected] wrote:
Yes. Actually, for reference it's best to look at refineMaxima(), which is the original we're working from. If that's not there then I can send a copy your way.
Super. I'll take a look and I'll open a new issue and update this one with progress.
Dr Rich Wareham