You are on page 1of 14

Implementation of 1-D

contourlet transform in
MATLAB M Jai Sidharth

BITS Pilani
Pilani Campus
Outline of presentation

• Features of contourlets
• Laplacian pyramid
• Directional filter bank
• Pseudocode
• Graphical results
• References

BITS Pilani, Pilani Campus


Features of Contourlet transform
• Multiresolution
• Perfect reconstruction
• Directionality
• Digital friendly nature
• Anisotropy

BITS Pilani, Pilani Campus


Laplacian pyramid
•of original
An iterative procedure that returns a downsampled bandpass version
  image.

In the decomposition level of the Laplacian pyramid -


• Convolve the output of previous stage with a lowpass filter W and
subsample it by two to create a reduced lowpass version.

• This image is then upsampled by inserting zeros in between each


row and column and interpolating the missing values by convolving
it with the same filter W to create the expanded lowpass image
which is subtracted pixel by pixel from the original to give the
bandpass image

BITS Pilani, Pilani Campus


Source: http://6.869.csail.mit.edu/fa16/lecture/lecture6.pdf BITS Pilani, Pilani Campus
Pyramid directional filter bank
Contourlet transform has a double filter bank structure
having two stages in cascade-
Laplacian pyramid- for multiscale decomposition
Directional filter bank- to capture directional information

BITS Pilani, Pilani Campus


Source: https://www.intechopen.com/media/chapter/41662/media/image2.png

BITS Pilani, Pilani Campus


Diagrams and graphs
• Graph – 1(Slide 6): analysis of performance with
respect to noise level and sampling rate

BITS Pilani, Pilani Campus


Pseudocode
•• Define
  parameters for filter bank and also decide
frequency, sampling rate etc.
• Generate the required signal and add gaussian noise to
it.
• Convert the signal to a 2-D array
• Apply Laplacian decomposition on it using lpdec and
obtain low-pass subsampled image ‘c’ and detail image
‘s’
• Apply directional filters to obtain all contourlet
coefficients in ‘y’. Corresponding to the level l used in
DFB we get images as output of convolution with
directional filters.
BITS Pilani, Pilani Campus
• Rearrange the coefficients into a vector ‘c’ and store
sizes of all subband images in a array ‘s’
• Find distribution of noise in contourlet domain by
decomposing image that contains only noise
• Find the threshold to eliminate the noise using results of
above step.
• Apply thresholding
• Reconstruct the signal using analysis filters
• Analyse performance using evaluation metrics
• Plot the results

BITS Pilani, Pilani Campus


Sampling Denoised signal
1x noise frequency Noisy signal SNR SNR RMSE

10100 17.34 18.06 0.8825

20100 17.01 21.21 0.61


1x noise 30000 16.4224 21.305 0.6083
implies that
noise added is
2x noise
1*randn()
10100 11.13 13.77 1.4405

20100 10.97 15.81 1.1989

30000 10.58 16.13 1.12

3x noise

10100 7.71 10.87 2.02

20100 7.34 10.81 2.03

30000 7.38 11.27 1.92

BITS Pilani, Pilani Campus


Analysis with F=1000Hz
before snr after snr rmse

25

21.21 21.31

20
18.06
17.34 17.01
16.42 16.13
15.81
15
13.77

11.13 10.97 10.87 11.27


10.58 10.81
10
7.71 7.34 7.38

2.02 2.03 1.92


1.44 1.2 1.12
0.88 0.61 0.61
0 1x noise

BITS Pilani, Pilani Campus


References
• Minh Do (2020). Contourlet toolbox
(https://www.mathworks.com/matlabcentral/fileexchange/8837
-contourlet-toolbox), MATLAB Central File Exchange.
Retrieved June 20, 2020.
• M. N. Do and M. Vetterli, "Contourlets: a directional
multiresolution image representation," Proceedings.
International Conference on Image Processing, Rochester,
NY, USA, 2002, pp. I-I, doi: 10.1109/ICIP.2002.1038034.

• R. Moore, S. Ezekiel and E. Blasch, "Denoising one-


dimensional signals with curvelets and contourlets," NAECON
2014 - IEEE National Aerospace and Electronics Conference,
Dayton, OH, 2014, pp. 189-194, doi:
10.1109/NAECON.2014.7045801.
BITS Pilani, Pilani Campus
Thank you

BITS Pilani, Pilani Campus

You might also like