Automatic number pate recognition system is a mass surveillance method that uses optical character recognition on images to read the license plates on vehicles. System might scan number plates at around one per second on cars traveling up to 100mph(160 km/h).they can use existing closed -circuit television or road-rule enforcement cameras, or ones specifically designed for the task. They are used by various police forces and as a method of electronic toll collection on pay-per-use roads and monitoring traffic activity, such as red light adherence in an intersection.
ANPR can be used to store the images capture by the cameras as well as the text from the license plate, with some configurable to store a photograph of the driver. Systems commonly use infrared lighting to allow the camera to take the picture at any time of the day. A powerful flash is inclined in at least one version of the intersection-monitoring cameras, serving both to illuminate the picture and to make the offender aware of his or her mistake. ANPR technology tends to be region -specific, owing to pate variation from place to place.
Some concerns about these systems have centered on privacy fears of government tracking citizens’ movements and media reports of misidentification and high error rates. However, as they have developed, the systems have become much more accurate and reliable. There is an increasing requirement to identify vehicles and track their location for a wide number of applications. These include:
Congestion charging – Several major cities around the world levy a charge a drive within them
Car park management – Using the number plate to identify the time of entry and departure of a
Counter-terrorism – Monitoring the arrival and departures of vehicles at major ports.
Our literature survey mainly focused on automatic number plate system research papers and its existing system along with its application, image processing technique and neural network recognition. These can be clearly illustrated as follows:
Automatic number plate recognition system
Javaanpr existing open source code in sourceforge.net
Thesis describing research, image processing and neural networking technique along with its algorithm in pdf on javaanpr on sourceforge.net
Image processing technique
ImageJ -api based on java language for digital image processing
Image editor -api based on java language made for image processing
JAI api -java advance imaging for image processing from sunmicrosystem at
Digital Image Processing (text book from library)
Neural networking technique
Introduction to java neural network second edition by jfheaton at heatonresearch.com
Some ocr samples using neuralnetworking at sourcecode.com and its explanation
Study on nepali ocr research conducted by madan puraskar guthi(yala Maya Kendra)
Ocr sample developed by Google based for Linux available for windows on dot net (tesseract)
Jooneengine-java api on neural network not so well developed and efficient at
Kohenen -java api on self organizing map applied to compress jpeg image.
Somdemo-sample java program for illustration how self organizing map works.
Program iterately train to converge with identical color from random samples
Artificial neural network text book available at library (low price edition from pearson
Neural networks systematic introduction by Raul Rojas(from lectures at free university at Berlin and later at the university of Halle)
Automatic Number Pate Recognition system
Javaanpr open source available at sourceforge.net worked as prototype for building our Nepali automatic Nepali number plate recognition system. It also contain thesis in pdf format prescribing image processing technique and neural networking technique along with its algorithm. It works well recognizing foreign number plates contained as sample in the site. It was beautifully coded applying sophisticated and specialized algorithms for image processing and neural network technique. It also used xml files to save and retrieve neural network training data.
Figure: sample javaanpr at sourceforge.net
For more information-http://sourceforge.net
Image processing Technique
ImageJ was first developed on class files now available on GUI interface. User can just process image using various buttons and entries if prescription is required .programmers can develop own macros and plugins to achieve its intended function if required and compile there within and run the code.
It is capable of processing both 2D and 3D interactive image processing.
Figure. ImageJ graphical window interface
For more information: http://rsb.info.nih.gov/ij/
b) Image editor
Image editor was also found during search for image processing tool. It is also based on java language and available as java API, now class file are available with GUI interface easing it’s its manipulation. Image editor api seems inefficient and not so capable for our intended operation and not so much researched.
C) JAI api
the java advance imaging(JAI) API further extends the java platforms (including the java 2D API) by allowing sophisticated, high -performance image processing to be incorporated into java applets and applications.JAI is a set of classes providing imaging functionality beyond that of Java 2D and the Java Foundation classes, though it is compatible with those APIs.
JAI implements a set of core image processing capabilities including image tiling, regions of interest, and deferred execution.JAI also offers a set of core image processing operators including many common point, area, and frequency-domain operators.
JAI is intended to meet the needs of all imaging applications. The API is highly extensible, allowing new image processing operations to be added in such a way as to appear to be a native part of it. Thus, JAI benefits virtually all Java developers who want to incorporate imaging into their applets and applications.
Flexible and Extensible
Initially program coding was done in JAI Later it becomes little inefficient and we again go for another programming method.
For further information-http://java.sun.com
d) Digital Image processing (text book from library)
The OpenCV implements a wide variety of tools for image interpretation. It is compatible with Intel® Image Processing Library (IPL) that implements low-level operations on digital images. In spite of primitives such as binarization, filtering, image statistics, pyramids, OpenCV is mostly a high-level library implementing algorithms for calibration techniques (Camera Calibration), feature detection (Feature) and tracking (Optical Flow),shape analysis(Geometry, Contour Processing ),motion analysis (Motion Templates, Estimators ), 3D reconstruction (View Morphing),object segmentation and recognition (Histogram, Embedded Hidden Markov Models, Eigen Objects).
The essential features of the library along with functionality and quality is performance. The algorithms are based on highly flexible data structures (Dynamic Data Structures) coupled with IPL data structures; more than a half of the functions have been assembler – optimized taking advantage of Intel® Architecture (Pentium®MMXâ„¢,Pentium® Pro, Pentium®III, Pentium®4).
Why We Need OpenCV Library
The OpenCV Library is a way of establishing an open source vision community that
Will make better use of up-to-date opportunities to apply computer vision in the
Growing PC environment. The software provides a set of image processing functions,
As well as image and pattern analysis functions. The functions are optimized for Intel®
Architecture processors, and are particularly effective at taking advantage of MMX››
The OpenCV Library has platform-independent interface and supplied with whole C
Sources. OpenCV is open.
Relation between Opens and Other Libraries
OpenCV is designed to be used together with Intel® Image Processing Library (IPL)
And extends the latter functionality toward image and pattern analysis. Therefore,
OpenCV shares the same image format (IplImage) with IPL.
Also, OpenCV uses Intel® Integrated Performance Primitives (IPP) on lower-level, if
It can locate the IPP binaries on startup.
IPP provides cross-platform interface to highly-optimized low-level functions that
Perform domain-specific operations, particularly, image processing and computer
Vision primitive operations. IPP exists on multiple platforms including IA32, IA64,
Source:-openCV reference manual.pdf
Cmgui-wx-2(.net wrapper class)
This openCV tool can be easily integrated with .net platform like c#, visual basic etc.
Cmgui is an advanced 3D visualization software package with modeling capabilities.Cmgui is a part of CMISS, a mathematical modeling environment initially developed by the University of Auckland Bioengineering Institute.CMISS stands for Continuum Mechanics, Image analysis. Signal processing and System Identification. There are three major CMISS software packages. Broadly speaking the main areas each piece of software deals with are as follows:
CM is used for computational modeling
Unemap is used for signal acquisition and processing
Cmgui is used for model visualization and manipulation
For more information:-wiki/getting started with cmgui
Neural Networking technique
a) Introduction to java neural network by jeff heaton
This book along with video lecture helped very much for us to understand neural networks and learn coding technique. It was published form Heaton research center and they have developed encog framework for neural network where programmer can build fast neural network prototype for fast testing and checking since easy and flexible. After parameters have been determined for best operation such as number of hidden layers and number of neurons in each layer coding can be done since it code will be inflexible for such modification. Book contained different chapters on various types of neural networks and also its application. Only first seven chapters are allowed to read online and rests are not. It provides all its source code on site which also helps in learning and testing.
Same book is also available in c# language.
For more information-http://heatonreasearch.com/
b) On the beginning of project research we also got OCR sample using neural network at sourcecode.com with explanation. It was written at c#, due to compiler problem I didn’t stress here much.
c) Nepali OCR
For us it was good news and opportunity to study research on Nepali OCR conducted by madan puraskar guthi. Different research papers were available on the site along with image processing portion code used to fragment Nepali character Image written on java. It deals with problem issues and complexity faced on Nepali character like devnagari font.
For more information -http://
d) OCR engine tessaract by Google
This was used by Nepali OCR for its processing and it supports many languages like Hindi, Nepali, Urdu, arabi etc. we didn’t research here much.
Figure: segmented portion of
Figure : Another segmented portion of
For more information-make Google search for link
d) joone engine
joone engine as a api in hope for easy and efficient coding we consider but it seems unworthy for project work. For beginner liking to test some xor operations and similar may find at least satisfactory otherwise unworthy.
For more information-http://www.jooneworld.com/docs/engine.html
This sample also seems beautiful in understanding self organizing map or kohenen network. Here it is used to compress jpeg image. It was programmed on seven packages.
For more information-http: //
f) som demo
This sample tries to converge iteratively with similar colors from randomly scattered pixel colors based on Euclidean distance method.
Figure: som before training
Figure: som after training
For more information-link available at reference http://www.ai-junkie.com/ann/som/
g) Artificial neural Network text book (library)
h) Neural network systematic introduction (by Raul Rojas)
This book is good for understanding neural network systematically and based on lectures at free university at Berlin and later at the University of Halle.
For more introduction-reference at http://www.wikipedia.com/selforganisingmap
Figure: sample kohenen neural network (3D kohenen feature map)
Anpr system application around world
On 11 March 2008, the Federal Constitution Court of Germany ruled that the laws permitting the use of automated number plate recognition systems in Germany violated te right to privacy.
Several Hungarian Auxiliary Police units use a system called Matrix Police in cooperation with the police. It consists of a portable computer equipped with a webcam that scans the stolen car database using automatic number plate recognition. The system is installed on the dashboard of selected patrol vehicles (PDA based handled versions exists as well) and is mainly used to control the license plate of parking cars, as the Auxiliary Police doesn’t have the authority to order moving vehicles to stop. If a stolen is found, the formal police are informed.
The UK has an extensive (ANPR) automatic number plate recognition CCTV network. Effectively, the police and security services track all car movements around the country and are able to track any car in close to real time. Vehicle movements are stored for 5 years in the National ANPR Data Centre to be analyzed for intelligence and to be used as evidence.
In the USA, ANPR systems are more commonly referred to as LPR (License Plate Reader or License Plate Recognition) technology or ALPR (Automatic License Plate Reader/Recognition) technology.
One of the biggest challenges with ALPR technology in the US is the accuracy of the Optical Character Recognition (OCR)-the actual identification of the characters on the license plate.
From time to time, states will make significant changes in their license plate protocol that will affect OCR accuracy. They may add a character or add a new license plate design. ALPR systems must adapt to these changes quickly in order to be effective.
In addition to the real-time processing of the license plate numbers, some ALPR systems in the US collect data at the time of each license plate capture .Data such as date and time stamps and GPS coordinates can be reviewed in relation to investigations and can help lead to critical breaks such as placing a suspect at a scene, witness identification, pattern recognition or the tracking of suspect individuals.
Average Speed cameras
Another use of ANPR in the UK, Italy and Dubai (UAE) is for speed cameras which work by tracking vehicles ‘ travel time between two fixed points ,and therefore calculate the average speed. These cameras are claimed to have an advantage over traditional speed cameras in maintaining steady legal speeds over extended distances, rather than encouraging heavy braking on approach to specific camera locations and subsequent acceleration back to illegal speeds.
The longest stretch of average speed cameras in the UK is found on the A77 road in Scotland, with 30 miles (48 km) being monitored between Glasgow and Ayr.
In Italian highways has developed a monitoring system named Tutor covering more than 1244 km (2007). Further extensions will add 900 km before the end of 2008.
The Tutor system is also able to intercept cars while changing lanes.
Many cities and district have developed traffic control systems to help the movement and flow of vehicles around the road network. This had topically involved looking at historical data, estimates, observations and statistics such as:
Car park usage
Pedestrian crossing usage
Number of vehicles along a road
Areas of low and high congestion
Frequency, location and cause of road words
The UK Company Traffic master has used ANPR since 1998 to estimate average traffic speeds on non-motorway roads without the results being skewed by local fluctuations caused by traffic lights and similar. The company now operates a network of over 4000 ANPR cameras ,but claims that only the four most central digits are identified , and no number plate data is retained.
Electronic toll collection
Ontario’s 407 ETR highway uses a combination of ANPR and radio transponders to toll vehicles entering and exiting the road. Radio antennas are located at each junction and detect the transponders, logging the unique identify of each vehicle in much the same way as the ANPR system does.
There are numerous other electronic toll collection networks which use combination of Radio frequency identification and ANPR. These include:
Bridge pass for the Saint John Harbor Bridge in Saint John New Brunswick
City link & Eastlink in Melbourne, Australia
Gateway Motorway and Logan Motorway, Brisbane , Australia
Fast Trak in California ,United states
Highway 6 in Israel
Tunnels in Hong Kong etc
Charge zones – the London congestion charge
The London congestion charge is an example of a system that charges motorists entering a payment area. Transport for London (TFL uses ANPR systems and charges motorists a daily fee of £8 paid before 10pm if they enter, leave or move around within the congestion charge zone.
Stockholm congestion tax
In Stockholm, Sweden, ANPR is used for the congestion tax of cars driving into or out of the inner city must pay a charge, depending on the time of the day.
ANPR systems may also be used for/by:
Section control, to measure average vehicle speed over longer distances.
Fillings stations to log when a motorist drives away without paying for their fuel.
A marketing tool to log patterns of use
Traffic management systems, which determine traffic flow using the time it takes vehicles to pass two ANPR sites.
Drive Through Customer Recognition, to automatically recognize customers based on their license plate and offer them their last selection, improving service to the customer
To assist visitor management systems in recognizing guest vehicles.
Circumvention Techniques (drawback)
Vehicles owners have used a variety of techniques in an attempt to evade ANPR systems and road -rule enforcement cameras in general. These methods may be
Increasing reflective properties of the lettering and so that system might no locate or produce high enough level of contrast to be able to read
Use of plate cover or spray
Use of dirt to smear their license plate or utilize covers to mask the plate
ANPR imaging hardware
The frontend of any Imaging hardware is image capturing device that is camera. Retroreflective camera returns the light back to the source and thus improves the contrast of the image. A camera that makes use of active infrared imaging (with a normal color filter over the lens and infrared illuminator next to it) benefits greatly from this as the infrared waves are reflected back from the plate. This is only possible on dedicated ANPR cameras, however, and so cameras used for other purposes must rely more heavily on the software capabilities.
Figure: hardware components used in ANPR system
To avoid blurring it is ideal to have the shutter speed of a dedicated camera set to 1/1000th of a second. License plate capture cameras can now produce usable images from vehicles traveling at 120 mph (190 km/h).threshold angles of incidence between camera lens and license plate are also major consideration to avoid image distortion during installation. Manufacturers have developed tools to eliminate errors from the physical installation of license plate capture cameras.
Research on down sampling character
For neural network input character image is down sampled into matrix whose value is binary 1 or 0 according to Boolean property of character on matrix region.
It showed that no of samples required is not fixed and it varies with thickness of font traced.
Figure: down sampling image character o with 7*5 matrix
Figure: downsampling same character image o (buffered) with 32 *35 matrix
Research works on algorithms
A new algorithm for character segmentation of license plate
Character segmentation is an important step in License Plate Recognition (LPR) system. There are many difficulties in this step, such as the influence of image noise, plate frame, rivet, the space mark, and so on. This new algorithm presents character segmentation using Hough transformation and the prior knowledge in horizontal and vertical segmentation respectively. Furthermore, a new object enhancement technique is used for image preprocessing. The experimentation results show a good performance of this new segmentation algorithm.
Determination of plate kind
Horizontal segmentation using Hough transformation
For more information:-a new algorithm for character segmentation of license plate.pdf
an adaptive thresholding algorithm for the augmented reality toolkit
It is well known that fixed global thresholds have adverse effects on the reliability of marker-based optical trackers under non-uniform lighting conditions. Mobile augmented reality applications, by their very nature, demand a certain level of robustness against varying external illumination from visual tracking algorithms currently AAR Toolkit depends on fixed-threshold image-binarization in order to detect candidate fiducials for further processing. In an effort to minimize tracking failure due to uniform shadows and reflections on a marker surface, a fast algorithm for selecting adaptive threshold values, based on the arithmetic mean of pixel intensities over a region-of- interest around candidate fiducials.
This works on a per-marker basis and evaluates the mean pixel luminance over a thresholding region-of -interest (ROI), which is defined as bounding rectangle around the marker’s axis -aligned corner vertices in screen space. If a marker has been detected in any given frame, its bounding rectangle will be used as thresholding -ROI prediction for successive frames. This method yields good thresholding level in practice, given sufficiently high video frame rates.
Fig.1.reflection off a marker’s surface with adaptive thresholding (upper) and a global threshold (lower)
For more information:-10.1.1.9.4636.pdf
adaptive license plate image extraction
This paper represents the automatic plate localization component of a car license plate recognition system. The approach concerns stages of preprocessing, edge detection, filtering, detection of the plate’s position, slope evaluation, and character segmentation and recognition. Single gray-level images are used as the only source of information. In the experiments Israeli and Bulgarian license plates were used, camera obtained at different daytime and whether conditions.
preprocessing for plate candidate identification
vertical edge detection
plate candidate segmentation
vertical projection acquisition
prime clipping of the plate
plate skew evaluation
plate candidate verification
Cray-level distribution consistency considerations
Cite This Work
To export a reference to this article please select a referencing stye below: