|
In this thesis, we develop an effective and fast automatic
vehicle license plate recognition system. From a digital vehicle
image without presetting the location and size of the license
plate, the proposed system can automatically find the location
of the plate, isolate the letters and digits occurring in the
plate, and then recognize the characters. The system includes
three parts: license plate location, character isolation, and
character recognition. Images with 256 gray levels are used in
this system. Firstly, vertical edges appearing in an image are
detected by the techniques of digital image processing. Since
the difference between the gray-levels of plate''s characters and
those of plate''s background are significantly, it is easy to
detect the edges of characters. By horizontal scanning, the
lines with enough gray-level changes are preserved. Then
possible locations of the plate could be identified from the
larger areas based on the intervals in the vertical and
horizontal projections. After locating the plate, the image of
the plate area is converted into binary for accelerating
recognition. Then the characters on the plate are isolated
according to the vertical projection of the above binary image.
Finally, the isolated characters are recognized by the moment
method, which compares each character with standard character
models by using the Euclidean distance. A postprocessing step is
executed if errors exist. The experiment was performed on a
Pentium personal computer. The experimental results show that 95
images whose license plates can be located correctly in 100 test
car images, and each image needs approximately 0.3 seconds for
locating the license plate.
|