Applied Signal
Processing
Projects to be handed in before February 15.
-
Recognizing number plates of cars using deconvolution
1. Make digital photographs of 10 different number plates at a distance of 1m. Add a sharp line to the photographs.
2. Measure the spatial resolution (smallest detail that can still berecognised, in pixels) of these number plates.
3. Simulate putting the number plates at successive larger distances, of 2, 5, 10, 20, 50, 100 etc. m,
which means degrading the resolution by factors of 2, 5, 10, 20 etc.
4. Now use various deconvolution methods to increase the spatial resolution. First use the methods discussed
in
exercise 2 of the lectures.
5. Now use the Maximum Entropy method (see
http://cmm.cit.nih.gov/maxent/letsgo.html).
6. Discuss the results. Up to which distance can number plates be recognised? Which methods work best?