Transcript License Plate Recognition (LPR)

By: Mani Baghaei Fard

During recent years number of moving
vehicles in roads and highways has been
considerably increased

Hence, Intelligent transportation systems (ITSs) have been
developed as a major tool for analyzing and also handling the
moving vehicles in cities and roads



These systems attempt to facilitate the problem of
identification of cars, via various techniques which
mainly rely on automated (rather than manual)
algorithms.
Image processing is one of these techniques
Unique property for identifying all vehicles is their
license plate number.

Security control of restricted areas

parking management systems

traffic law enforcements

surveillance systems

Electronic toll collection


Difficulties:
Poor image resolution usually because the plate is too far away but sometimes resulting
from the use of a low-quality camera.

Weather condition

complex background

plate deficiencies (damaged or dirty)

Supporting specific range of distances

Viewpoints

Blurry images, particularly motion blur

Poor lighting and low contrast due to overexposure, reflection or shadows


response time is another restriction in real time applications such as license plate
tracking
Lack of coordination between countries or states which results in different design of the
plates.




A license plate recognition (LPR) system
mainly consists of three major parts
license plate detection (LPD)
character segmentation
and Optical Character Recognition(OCR)


The task of recognizing specific object (i.e.
Car license plate here) in an image is one of
the most difficult topics in the field of
computer vision
There are many methods



Edge-based techniques:
methods based on edge analysis combined with morphology
operations achieved promising results . Presence of dark
characters on the light background at license plate provides
strong edges which can be used as a cue to detect the license
plate.
Unfortunately, solely using edge information, fails the
algorithm in complex scenes. Hence, combining edge
information with other cues improves the detection rate.

Hough transform: attempts to find the
rectangular shapes.
Advantages: useful in finding the boundary box
of a license plate regardless of characters.
Disadvantages:
Not suitable for distorted or dirty plate
Computational complexity
is only suited for closed shut





Texture analysis: This approach takes the advantage of
existing homogenous and frequent texture-like edges in
the plate region.
Gabor filters have been one of the major tools for texture
analysis
Using these filters, the process is independent of rotation and
scaling. It has the ability of studying images in an unlimited
number of directions. But it is a time consuming
and complex method specially when applied to large images.






Fuzzy Logic-based
Texture – based
Neural networks
Train and test techniques such as Adaboost
And many others….
Definitely ,I am not going to details about all
of them …!!!

By observing license plates in images, two main features are
noticed:
horizontal edges around a car
plate are relatively strong and
dominant.
1)
density of vertical edges
across a car plate are significant.
1)
RGB
Gray
Vertical & Horizontal Edges
Horizontal Edges
Vertical Edges

Roberts

Log

Zero cross

Canny

Prewitt

Sobel
Sobel Operator is a [1 2 1] filter
Can find Vertical and
Horizontal edges
seprately
By experience Sobel
Preferred Cause of
better response
a major cause of failure for a plate detection
system is low quality of car image. In order to
improve the quality of plate image I used a
pre-processing algorithm which increases the
image contrast at locations where might be a
license plate.


variance of local intensity for constituting
pixels of the license plate has a limited range
and does not change dynamically. This
function increases image


Based on some experiments the local
intensity variance for a plate region can
be out of considered range 0–60.
method does not work well under severe
illumination change.

He replaced the variance of image intensity
with the density of vertical edges in Zheng`s
method







1-Reading image
2-RGB2gray
3-find out vertical edges using Sobel operator
4-Dilation along X axes
5-smoothing
6-using morphological tools to extract plate
7-enhancement and plate preparation for OCR
algorithm