An efficient method of license plate location

1
An efficient method of license plate location

• Pattern Recognition Letters 26 (2005) 2431-2438
• Journal of Electronic Imaging 11(4), 507-516
  (October 2002)

Presented by - Waseem Khatri
2
Objective

• To efficiently locate a license plate in an image
  
• Motivation
• License plate recognition can be an essential
  tool for
• Road traffic monitoring
• Automatic payments of tolls on highways bridges
• Parking lot access control
• Ticketing speeding vehicles

3
Algorithm
Image
Enhancement
Vertical Edge Extraction
Noise Removal
Plate Location
License plate Extraction from Original image
4
Edge Information

• Plate area contains rich edge information
• Background areas around the plate mainly include
  horizontal edges
• Background areas have long curves and random
  noises
• If only the vertical edges are extracted from the
  car image and most of the background is removed,
  the plate area can be isolated

5
Image Enhancement

• The input image is converted to a gray scale
  image
• of size 384 X 288
• Gradients in the image due to improper lighting
  conditions
• Few vertical edges in the plate area
• Enhancement is necessary
• Calculate the luminance and variance of each
  pixel
• Bilinear Interpolation

Enhancement Coefficient
8 X 8 Blocks
6
Edge Extraction and Noise Removal

• Vertical edge extraction using Sobel Operator
• Thresholding
• Background curve and noise removal is done using
  the Concerned Neighborhood Pixel (CNP) Algorithm
• CNP checks all pixels around the concerned pixel
  and decides if its a randam noise pixel or a
  genuine edge pixel

7
License plate search

• A window of size (H X W) is passed through the
  CNP output image
• Total number of edge points in the window are
  counted
• Candidates are selected if they are above a
  certain threshold
• Maximum value among the candidates is considered
  as a final result
• The co-ordinates are noted and the plate is
  extracted from the orignal image

8
Results
9
Results
10
System Application
Enhancement BLI
Vertical Edge Extraction
Noise Removal CNP
Plate Location
Image
Affine Transformation
Output System
Character Extraction (Segmentation)
Classifier
Hotelling Transform
Blob Coloring
Bayesian
Fisher
Neural Nets
11
Conclusion

• Advantages
• Higher recognition rate compared to other methods
  like Line sensitive filters (Luis et al., 1999),
  Row-wise Column-wise DFTs (Parisi et al.,
  1998), Edge image improvement method (Ming et
  al., 1996)
• Drawbacks
• Slower than other four methods
• Calculated values of luminance and variance using
  Bilinear Interpolation are not actual values
• Image size is fixed 288 X 384
• Window size of the license plate search is fixed
• Bilinear Interpolation is the most
  computationally intensive procedure in the given
  algorithm