possible bug in convol_fft

[markus.sc...@gmail.com]

Hi, 

in IDL 8.7 I have spotted possible bugs in convol_fft. While the output results are correct, it sometimes needlessly repeats FFTs. The way it is written, if image_FFT or kernel_FFT are provided and have the right sizes, they are not recomputed. But these size tests needlessly fail, if the inputs are not square or in the case where /no_padding is set and the kernel has a different size than the image. 

For the /no_padding case, the correct test should be

MAX(imageSz.DIMENSIONS[0:1] ne imageDims) ;removed [0]

MAX(kernelSz.DIMENSIONS[0:1] ne imageDims);replaced kernelDims[0]

and the other case

MAX(imageSz.DIMENSIONS[0:1] ne imageDims*2) ;removed [0]

MAX(kernelSz.DIMENSIONS[0:1] ne imageDims*2) ;removed [0]

Furthermore, the documentation for image_FFT and kernel_FFT are not clear. While they correctly instruct to set them as returned from previous calls to convol_FFT, and correctly state what they return, it does not mention that setting them avoids recomputing these FFTs.

Can someone with a current version of IDL check, if this problem is still around? 

You can find it in lib/convol_fft.pro, probably lines 73,86,113 & 130.

Thanks, Markus

[Chris Torrence]

Hi Markus,

Thanks for reporting this. It is still a problem in IDL 9.1. Luckily, as you point out, the results are correct, it is just less efficient than it could be.

I have gone ahead and added all your fixes and documentation changes for IDL 9.2, shipping sometime in the next few months.

Thanks again,

Chris

NV5 Geospatial Software