I am working on image processing by a 3D wavelet transform.

I have a problem with the classification size of the wavelet coefficients. As you know, when we apply 3D_DWT on the image (for example: Hyperion of the area by 256*256*128 size) the sub-band size would be decreased (for LLL in 1l level the size is 128*128*64).

How I could apply down sampling just for the Z direction (spectral direction)?

It means this size of the sub-band should be 256*256*62 after one level wavelet applying. I want to do it in the Matlab script (dwt3.m & wavedec3.m), but when I removed the down sampling, it was applied in all three directions.


1 Answer 1


You are applying a 3D DWT scheme whose down-sampling is coherent across the three directions. Nothing forbids you from decomposing bands the way you want, and the literature is full of so-called isotropic, Tensor, Separable, Square or Standard implementations (vocabulary: separable/non-separable, standard/non-standard, S-form/NS-form, rectangular/square, anisotropic/isotropic, tensor/Mallat, hyperbolic/isotropic, separated/combined wavelet transform)

To do that, it is more appropriate to perform 1D wavedec passes your own way on each direction separately.

  • $\begingroup$ Thanks for your replying, I will try to apply 1D_DWT on each direction and for 4 level, it should be repeated for LLL sub bands. $\endgroup$
    – Mehdi
    Nov 2, 2017 at 14:47

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.