Super Resolution

Super Resolution is a new field in forensic science. Research reports (see eg SPIE-references) indicate that this is a challenging way in improving the quality in forensic science.

We state that there should be a quality control in Super Resolution. We have tested Dutch number plates in a sequence. The camera used was a Sony Digital Handycam.

We reduced the resolution of this sequence with 10 percent (selecting 1 out of 10 lines) :

Then we calculated the restored image :

Please note that the BP-64-XH is now changed to DP-64-XH.

The next procedure was used for getting this result :

•

•Frame splitting
•Magnify the images with an interpolation method which takes sub pixel resolution into account
•Register the images.
•Average the images

Similar results were obtained with commercial packages.

For blurred images it is much harder to find the restore the image. Please try for yourself with Dutch Number Plates and zoom them. I have a sequence available on line. The software that can be used with Khoros is also available here with some examples of other sequences of license plates.

Discussion

The reason that this error was obtained is two-fold :

- the compression that is used in the Sony camera
- the stroke in the B is filtered by this method of super resolution. We will recognize the D in the image, because the restored image seems to be clear.

This example shows that it is important to validate this new method of image processing. Furthermore, compression that is used in digital data streams might influence the results that are obtained.

If you would like to participate in the discussion on validating super resolution, please find this discussion group on line. If you have any questions, please contact Zeno Geradts.