Vehicle License Plate (VLP) Recognition System By German H. Flores and Gurpal Bhoot Agenda Introduction Goal and Motivation Image Segmentation Feature Extraction Classification Results/Conclusion Future Work Introduction Technological advancements in both software and hardware Better ways to capture, edit and analyze images Safety and security of pedestrians and people in motorized vehicles The large number of cars on the roads has increased the probability of an accident occurring With a VLP system, the owner of a car can be easily identified and held responsible for their actions Video Process Flow •Ex: Separate LP from car and background as well as characters from LP •Extract features that can be used for classification •Ex: Area, Perimeter, Number of Corners, Contains Hole Pattern Classification •Locate objects and boundaries in images Feature Extraction Image Segmentation Object Recognition Process •Take the features extracted from the image and use them to automatically classify image objects •Ex: Classify either as letters (A-Z) and/or numbers (0-9) Assumptions Ideal lighting Conditions Non-white car License Plate is in the same region License Plates are similar sizes Only California license plates after 1987 License Plates must be white with dark characters Upper case letter O and 0 are the same Image Segmentation Binary Image Convert image the original image into a binary Threshold was chosen through testing Binary Image Resize Image Shrink the image Cut out the background Leave only part of the image where license plate is most likely to appear Image Segmentation Windowing Method Windowing Method used to find the license plate from the binary image Send a window (m X n) through binary image, pixel by pixel Resized Binary Image Image Segmentation Windowing Method Find the license plate by number of white pixels Below is the resulting image from applying the Window Method Final Binary Image Image Segmentation Connected Component Algorithm Used for separating license plate from the image Finds the different objects Finds the license plate by size and shape Extracted License Plate Then used for separating the letters and numbers Finds each character and extracts them one by one Image Segmentation Feature Extraction What features are important for a successful pattern classification? Ex: Color, Area, Perimeter, mean, variance Character Area Recognition Perimeter Number of Corners in compressed simple image Compressed and Perimeter of Has Hole Normalized Contour Character Image Number of Corners in Distance compresse Image d full image Feature Extraction Area Simple Compression And Normalized Corners Perimeter Compressed and Normalized Full Compression And Perimeter of Contour Normalized Corners Feature Extraction (http://www.leewardpro.com/articles/licplatefonts/font-penitentiary.html) Characters that have holes ABDOPQR0 689 Characters that do not have holes CEFGHIJKL MNSTUVWX YZ123457 Features: Area Perimeter Perimeter of Contour Number of Corners in simple compressed Image • Number of Corners in full compressed Image • • • • • Distance Image • Normalized Character Image Feature Extraction Harris Corner Detection A corner can be defined as the intersection of two edges A new Corner Matching Algorithm Based on Gradient. (Yu, Haliyan.,., Ren Cuihua., and Qiao Xiaoling) Feature Extraction Feature Extraction 1. Compute X and Y derivatives of the grayscale image Gx Gy 2. Compute products of derivatives 3. Define at each pixel (x,y), the matrix 4. Compute the response at each pixel 5. Threshold on Value R 0s or negative numbers are the corners Feature Extraction CHARACTER AREA PERIMETER A B C D E 103 120 95 117 90 74 106 75 99 86 HAS HOLES 1 1 0 1 0 Character Features Extracted From Image PERIMETER OF CONTOUR 85 102 70 81 50 Number of Corners in simple compression 63 51 63 43 36 Number of Corners in full compression 202 262 255 270 438 Character Features from Database Correlation Corr2() Results LICENSE PLATE LICENSE PLATE CHARACTERS RECOGNIZED 3DDF536 -- D 5 3 EZEZBEH E 2 E Z B E 3HOS909 H O 9 3 S 9 0 4HCF116 4 H C F 1 1 6 2LOX542 2 O X 5 4 2 4FJF892 4 F F 8 9 2 J 3TFB805 T F B 3 8 0 5 3WVD539 3 3 9 3GXP106 3 G X P 1 O 6 4EYB802 4 E Y B 8 0 2 4DNX245 --- 4 D N X 2 4 5 4CGS613 --- C G S 6 1 3 --- 3XHK859 3 X H X 8 5 9 3JXK363 X K 6 3 Results Results Conclusion/Overview Raw Image Image •License Segmentation Plate Letter Segmentation •Characters Feature Extraction ABDOP QR0689 CEFGHI JKLMNS TUVWXY Z123457 •Area •Perimeter •Number of Corners Character Feature Database •All the characters (A-Z) and (0-9) Classification •Correlation Bibliography