Home My Page Projects MPTK: The Matching Pursuit ToolKit
Summary Activity Forums Tracker Lists Docs News SCM Files

Forum: help

Monitor Forum | Start New Thread Start New Thread
RE: MPTK Wavelet [ Reply ]
By: Rémi Gribonval on 2007-11-23 15:48
[forum:5420]
Dear Frederic,

The anywave block can be used with different scales too, but you have to specify the waveform at each scale. To perform a wavelet decomposition with anywave blocks, you would have to build a dictionary by concatenating as many blocks as the number of wavelet scales you want and assign the waveform of one specific wavelet scale to each block.

I am afraid this would be far less efficient than to use a wavelet transform, so if you want to use wavelets with MPTK, you'd better write a new wavelet block, using the new block/atom plugin API which was just released in MPTK 0.5.4

I think a wavelet block would definitely make a lot of sense and many other MPTK users would be interested, so it is worth keeping this discussion on an MPTK forum so that other users / developpers can contribute. May I suggest that you subscribe to the mptk-devel mailing list? This is the place to discuss implementation issues, and we strongly encourage developers to use it to share their experience developing with MPTK. It will also help us keep you informed about / get your opinion about future evolutions of MPTK.

As a starting point, I also suggest that you have a look at existing blocks/atoms and templates (in version 0.5.4, which implements a new plugin API) and come back with more specific questions if you meet problems. We are working hard on improving the robustness, speed and flexibility of the core MPTK library, which unfortunately leaves us limited time for help, but we will certainly do our best, and again, I think that a wavelet block is a highly desirable feature.

For the MPTK development team,

Remi Gribonval.

MPTK Wavelet [ Reply ]
By: Frédéric CAUDAL on 2007-11-23 14:36
[forum:5419]
Hello,

I tried to use the "anywave block" to create a daubechies wavelet but as a result on the signal, the wavelet is considered as a fixed signature and not a wavelet. There is no dilatation to match the signal like the gabor block and others.
Could you tell me how to create a new wavelet block and not a signature matching?

Thank you

Frédéric