Pyramid Coder with Nonlinear Prediction - PowerPoint PPT Presentation

About This Presentation
Title:

Pyramid Coder with Nonlinear Prediction

Description:

Pyramid Coder with Nonlinear Prediction. Burt/Adelson pyramid coder. A nonlinear prediction ... Burt/Adelson coder. with nonlinear prediction. Aliasing Effect ... – PowerPoint PPT presentation

Number of Views:27
Avg rating:3.0/5.0
Slides: 14
Provided by: Panu
Learn more at: https://web.stanford.edu
Category:

less

Transcript and Presenter's Notes

Title: Pyramid Coder with Nonlinear Prediction


1
Pyramid Coder with Nonlinear Prediction
Panu Chaichanavong
  • Burt/Adelson pyramid coder
  • A nonlinear prediction
  • Aliasing effect
  • A switching method
  • Conclusion

2
Burt/Adelson Pyramid Coder
Introduced by Burt and Adelson (1983)
Gaussian
Laplacian
Quantized
and transmitted
Original image
Reconstructed image
3
A Nonlinear Prediction
By Florencio and Schafer (1994)
Filter Just subsample!
Interpolator
  1. Replication
  2. Weighted median of 6 neighbor pixels
  3. Median of 4 neighbor pixels
  • 1 1
  • 1
  • 1 1

1 2 1 2 3 2 1 2 1 2 3 2 1 2 1
Lower resolution image
Predicted image
4
A Nonlinear Prediction (2)
Burt/Adelson coder
with nonlinear prediction
5
Aliasing Effect
Test image
6
Aliasing Effect (2)
Gaussian filter
Only subsampling
7
A Switching Method
Lets look at the following 4x4 filters
.032 .058 .058 .032 .058 .102 .102
.058 .058 .102 .102 .058 .032 .058 .058 .032
0 0 0 0 0 0 0 0 0 0 1 0 0 0
0 0
0 0 0 0 0 1 2 0 0 3 4 0 0 0
0 0
Filter3 Linear low-pass with weight shown above
Filter2 Median of average
Filter1 Pick one !
8
A Switching Method (2)
Input1
Input2
Input1 may be an edge, use filter1 Input2
high frequency, use filter3
How can we distinguish these inputs?
9
A Switching Method (3)
Let D sum of difference of adjacent pixel
values sd standard deviation of 16 pixel
values p D/sd
Decision criterion p lt 18, use filter1 18 lt p lt
22, use filter2 p gt 22, use filter3
10
A Switching Method (4)
Filtered subsampled image
Filter used
Filter1
Filter2
Filter3
11
A Switching Method (5)
Method PSNR (dB) Bit-rate (bit/pixel)
B/A 32.77 0.8278
Nonlinear 34.00 0.6544
Switch 33.58 0.6501
Reconstructed image
12
A Switching Method (6)
Images in the Gaussian pyramid
Filter used
Filter1
Filter2
Filter3
13
Conclusion
  • Low-pass filter reduces aliasing effect but gives
    blurred image
  • Some nonlinear prediction preserves edges and
    details but may introduces annoying aliasing
  • A decision criterion is presented to switch among
    various filters to select an appropriate one for
    a particular input
Write a Comment
User Comments (0)
About PowerShow.com