Exercise 4

  1. These are the starting images:
    lena.png lena512-Gauss-GaussianNoise.png
    The first image is a standard test image.   The second image was artificially produced by someone else ((Lena Blurred and Noisy) http://telin.ugent.be/~frooms/friep/screenshots/), presumably from the first image or one like it.   The goals are to estimate the convolution kernel and use it to deconvolve the second image.  

  2. Open the first image.

  3. Change color space to YCbCr by clicking CSpace>YCbCr.

  4. Make only the Y channel active by pressing keypad 5 and 9.   The status should be displayed in the 3rd pane of the status bar.

  5. Press X to create second layer.

  6. Open the second image.

  7. Click Blur/Sharpen>Find Kernel from Blurred and Sharp Images.  

  8. In the following Dialog Box, which automatically apears, make these settings:
    and press OK.

  9. Click Blur/Sharpen>View Kernel and, in the dialog box, set View Kernel to False and View Singular Value Decomposition (SVD) to True.   After clicking OK, the following should open in Window's Notepad:

    Note in row following "7, 1," only the first value is relatively large.   That indicates that we can approximate the kernel to increase the speed of deconvolution.   Close Note Pad.

  10. Click Blur/Sharpen>Deconvolution by PseudoLAD and make these entries in the dialog box:

  11. After clicking OK, the image should look like this:
    and note the PSNR, 32.190266, shown in the 5th pane of the status bar. Note that regions of the width of the kernel, in this case 7 pixels, next to the image borders, are ignored in this PSNR calculation, in order to ignore unavoidable artifacts in this region.  

  12. Compare with results published at http://telin.ugent.be/~frooms/friep/screenshots/ For example, to get PSNR of lena512-Gauss-GaussianNoise-RichardsonLucy-TV.png, open this image and leave the first image in the bottom layer.   Click Blur/Sharpen>Deconvolution by PseudoLAD and set the Outer Iterations, on line 12, to 0 which will turn the deconvolution off.   After clicking OK, the PSNR will appear in the status bar.   I get 31.712722 .

  13. Finished!