Automatic License Plate Location Using Template Matching

Kerry Widder

This project explored the use of template matching for the automatic
location of license plates in the images of vehicles.  The template
consists of the image of a license plate.  Two methods will be explored
as candidates for determining the best match of the template in a given
image.  The first method will use correlation between the sobel filtered
image and template.  The second method will use the invariant moment
descriptors as a measure of the degree of similarity between the
template and the objects in the image.  The objects in the image will be
identified by finding the boundaries in the sobel filtered image. 

The methods were applied to ten sample vehicle images.  The correlation
method showed modest success (60%).  The moment invariant method was
less successful (20%) with a major problem being the difficulty of
isolating the boundary of the license plate from other features of the
vehicles.  The scope of the project was limited to a small sample size
and relatively ideal conditions.