Automatic License Plate Location Using Template Matching

1

• Automatic License Plate Location Using Template
  Matching
• University of Wisconsin - Madison
• ECE 533 Image Processing
• Fall 2004 Project
• Kerry Widder

2
Problem Statement
What Automatically locate a license plate in the
image of a vehicle Who Law enforcement, parking
structures Why Quicker, cheaper Difficulties
variations lighting, angle, size, location,
distance, color, vehicle features, contrast,
holders
3
Approaches
Distance sets (spatial arrangement of
features) Morphological operations Contrast
levels Contrast transitions row, column Genetic
algorithms Pattern matching
4
Approach
Approach used in this project template
matching Two methods used to measure a template
match 1. Correlation The greater the
similarity between the template and the image in
a particular location, the greater the value
resulting from the correlation. 2. Moment
Invariants A set of regional statistical
descriptors invariant to translation, rotation
and scale changes.
5
Implementation Algorithm
Filter image (gaussian, then Sobel to find
edges) Calculate correlation and correlation on
correlation Threshold sobel image to convert to
binary Perform Morphological opening to reduce
protrusions Find boundaries in binary
image Calculate moment invariants of each object
and compare to template to find best match
6
Implementation Data
Obtained ten images of vehicles Conditions were
controlled sunny, same distance All images were
processed through the algorithm implemented
in MATLAB
7
Implementation Data sample of processed data
Input image
Edge image
Correlation result
Boundaries
8
Implementation Data Templates
License plate template Correlation
template
9
Results Correlation - success
Sobel image Best
matches marked
Correlation result
Correlation on correlation
10
Results Correlation - failure
Correlation image
Best matches marked
Correlation image Best
matches marked
11
Results Moment Invariants - success
Input image
Boundaries
Selected match
12
Results Moment Invariants - failure
Input image
Boundaries - insufficient
Input image
Boundaries - protrusions
13
Results Summary
Correlation 60 success rate Correlation on
correlation 20 success rate Moment
Invariants 20 success rate
14
Discussion
Correlation moderate success, correlation on
correlation not better Further work refine
template? (will be difficult to do for all
cases) Moment Invariants not successful Further
work refine boundary/object identification
(difficult due to wide variations in vehicle
features, lighting, etc.) Limitations sample
size small, images controlled
15
Conclusion
Simple template matching, using correlation or
moment invariants, does not appear to be a good
candidate for automatic license plate location.