车牌识别

VEHICLELICENSEPLATERECOGNITIONUSING

MORPHOLOGY ANDNEURALNETWORK

Sneha G. Patel

Sardar Vallabhbhai Patel Institute of Technology, Vasad, 388 306, Gujarat, India

snehapatel11@gmail.com

ABSTRACT

Automatic Identification of vehicles is a verychallenging area, which is in contrast to the traditionalpractice of monitoring the vehicles manually. Automatic license plate (LP) recognition is one of the mostpromising aspects of applying computer vision techniques towards intelligent transportationsystem. InLocation of the vehicle plate, a method of vehicle license plate character segmentation and extractionbased on improved edge detection and Mathematical morphology was presented. In the first place, colorimages were changed into grey images, secondly through calculates the difference of each pixel andneighbourhood pixels to build up images edge and it can make the license plate stand out; Sobel operatoris used to extract the edge of objects in image; then the algorithm applies the dilation anderosionmathematical morphology of binary images to get the image smooth contour.The segmentation resultwhich is sent forward to LP recognition stage will improve further processing’s efficiency. Neural Networkis used to recognize the license plate character. Because of the accuracy of the plate region extraction, thecharacter can be extracted exactly from the plate region.

KEYWORDS

License Plate Recognition, Back propagation, Neural Network,Median Filter.

1.INTRODUCTION

While the first industrial automatic system for Car License Plate Recognition (LPR) wasintroduced in the 80’s, an outburst of commercial systems occurred in the 90s. Although that a lotof LPR systems are available in the market, the research and development still continues and newsophisticated solutions to plate localization, character segmentation and recognitionappear.Vehicle’s license plate recognition system has been a special area of interest in videosurveillance area for more than a decade or so. With the advent of sophisticated video vehicledetection systems for traffic management applications, number plate recognition system findswide varieties of places to fit itself beyond just controlling access to a tollcollection point orparking lot [3]. It can now be integrated to the video vehicle detection systems which usually areinstalled in places of interest for intersection control, traffic monitoring etc., to identify vehiclethat violates traffic laws or tofind stolen vehicles. This paper presents a method of vehicle licenseplate Character recognition. The whole system into three following steps:

1-Plate location or finding location of plate in the vehicle image and cropping plate image fromit.

2-Platesegmentation or cutting plate image to character’s images.

3-Character recognition or convert character’s images to final distinguished characters amongthem.

Due to the diversity of parameters involved in car images,License plate detection is consideredthe most crucial stage in the whole LPR system. In the past, a number of techniques have beenproposed for locating the desired plate through visual image processing. Now there are somealgorithms about the locating oflicense plate, such us the methods based on color feature [4],

DOI : 10.5121/ijci.2013.2101

1

International Journal on Cybernetics & Informatics ( IJCI) Vol.2, No.1, February 2013

edge extracting [6-8], histogram analysis [9], symmetry [11], morphological operators [12],.In amethod of Chinese license plate recognition has been proposed that structural verification is usedfor plate locating and projection segmentation is exploited for incising plate image intocharacter’s images.Sobel operator is the first and best-known choice for most LPD algorithms toconvert the grayscale scene image to the gradient image at the initial stage.[12] In general,conventional LPD algorithms focus only on vertical edge detection through vertical Sobeloperator, because the horizontal edge density of license plates is not more outstanding than that ofthe surrounding background objects. Character recognition of the number-plate is a fairly welldeveloped field in computer vision in which template matching and neural networks are oftenused and can produce satisfactory results [14-15]. However, template matching has its drawbacksin some aspects comparing with neural networks. For example, when characters of number-plateare segmented, neural network approach is preferred to template match one. The paper isorganized as following way. Section I is of introduction of whole system.We shall introduce theproposed recognition algorithm in section II to V. Experiment and Conclusion will be shown inSection VI and Section VII respectively.

2.PRE-PROCESSING

Preprocessing mostly is necessary to facilitate further high performance recognition, in thisproposed methodology, the character is binarized and the noise is eliminated in the preprocessingstage.

2.1.Grey Conversion

We have taken color image pf car clearly showing its License plate for experiment.We firstlyconvert this RGB color input image to a 256 grayscale image usingformula (1).GREY=J(:,:,1)+J(:,:,2)+J(:,:,3)………….(1)

Figure 1. Original color image

The grey-scale could take every pixel of the picture to a number between 0-255 and the purposeof the binarization is to take every pixel into the number 0 or 255.To remove the tonal variationbetween Red, Green and Blue channels of input images and converting it into grey scale flatnessto a single hue.

2

International Journal on Cybernetics & Informatics ( IJCI) Vol.2, No.1, February 2013

Figure 2. Grey image

2.2.MedianFiltering

It is inevitable for containing noises of original image. We use median filtering to eliminate thenoises. Using median filtering not only can eliminate the noises, but also make the high frequencymore concentrated. There by, it is beneficial for us to detect the edges in images.The salt andpepper characteristic of vehicle license plate is presented, So Median filter is best known choicefor removing such noise.

3.LICENSEPLATEEXTRACTION

To extract the license plate, we mainly utilize the characteristic of the plate in vehicle images likethe color in the region is blue. We project the binary difference image horizontally and vertically,properly smooth these two projection curves, and search their peaks to find the accurate rectanglethat includes the plate.[7]

Figure.3. Thresholded Binary image

3.1Opening and Closing

Opening generally smoothes the contour of an object, breaks narrow isthmuses and eliminatesthin protrusions.Closingalso tends to smooth sections of contours but, as opposed to opening, itgenerally fuses narrow breaks and long thin gulfs, eliminates small holes, and fills gaps in thecontour [10].

The opening of setAby structuring elementB, denotedA_B, is defined asA_B= (A_B)⊕B(2)

Thus, the opening ofAbyBis the erosion ofAbyB,followed by a dilation of the result byB.Similarly, closing is defined asA•B= (A⊕B)_B(3)

3

International Journal on Cybernetics & Informatics ( IJCI) Vol.2, No.1, February 2013

Figure 4.Opened image

This says that the closing ofAbyBis simply the dilation ofAbyBfollowed by the erosion of theresult byB.

4.CHARACTERSEGMENTATION

Character segmentation is an important stage in many license plate recognition systems. There aremany factors that cause the character segmentation task difficult, such as image noise, plateframe, rivet, and rotation and illumination variance [13].Object segmentation is an essential taskin computer vision and object recognitions. Image segmentation is the process of partitioning adigital image into multiple regions or sets of pixels [6]. These partitions represent differentobjects in the image, usually having the same texture or color. Segmentation is quite essential toimage feature extraction and subsequent classification of the resultant features.This step is verysignificant due to overlapping characters that form the license plate. There are threemain formsof characters that are overlapping vertically, ligature, diacritics, horizontal overlap, and twoconnected characters. The task will be more difficult for those different forms ofWhich are joined.

Figure 5. Segmented characters of Licenseplate.

5.CHARACTERRECOGNITION

The two methods are template-matching method and neural network method.Here we areevaluating neural network method.Character recognition final step in vehicle license platedetection and recognition is reading of singlecharacters and numbers. This step is very importantfor example at the entrance to car-park or for the police for stolen cars search. Single elements onlicense plate must be segmented and analyzed. The analysis is called as Optical CharacterRecognition(OCR) using ANN [13].

4

International Journal on Cybernetics & Informatics ( IJCI) Vol.2, No.1, February 2013

5.1 Artificial Neural Networks

Artificial neural networks are statistical models of real world systems which are built by tuning aset of parameters. These parameters, known as weights, describe a model which forms a mappingfrom aset of given values known as inputs to an associated set of values, the outputs. The processof tuning the weights to the correct values–training-is vehicleried out by passing a set ofexamples of input-output pairs through the model and adjusting the weights in order to minimizethe error between the answer the network gives and the desired output. Once the weights havebeen set, the model is able to produce answers for input values which were not included in thetraining data. The models do not refer to the training data after they have been trained; in thissense they are a functional summary of the training data [1,10].

The use of artificial neural network (ANN) in OCR applications can dramatically simplify thecode and improve quality of recognition while achieving good performance. Another benefit ofusing neural network in OCR is extensibility of the system i.e. ability to recognize more charactersets than initially defined. In the character recognition algorithm using neural networks, theweights of the neural network were adjusted by training it using back propagation algorithm. Thesize of each character is 28-by-18pixels which are arranged column wise to give 504 _ 1 arraysas input. In order to train the neural network, we have created different sets each containing digitsfrom 0 to 9. This is called Block training. The complete net work was implemented as a library,which was statically tied to the project. This helped to isolate the neural network code from therest of the preprocessing and segmentation code [14]. It also helped to reduce the memoryrequired for the program.Character recognition of the number-plate is a fairly well developedfield in computer vision in which template matching and neural networks are often used and canproduce satisfactory resultsHowever, template matching has its drawbacks in some aspectscomparing with neural networks. For example, when characters of number-plate are segmented,neural network approach is preferred to template match one due to more computation cost of thelater method [6].

As a very powerful technique for pattern recognition problems, it is not surprising that neuralnetworks became the basis of a number of such systems [15].

5.2 The Back Propagation (Bp) Algorithm

The back propagation is a widely used algorithm, and it can map non-linear processes. It is a feedforward network with the one or more hidden layers. The elementary architecture of the backpropagation network has three layers. There are no constraints about the number of hidden layers.Back propagation is a systematic method for training multilayer artificial neural net works. It hasa mathematical foundation that is strong if not highly practical.

Using neural network has advantage from existing correlation and statistics template techniques[5] that allow being stable to noises and some position modifications of characters on licenseplate. The network must memorize all the Training Data (36 characters). For the validation of thenetwork we have built a program that reads the sequence of characters, to cut each character andresize it and put the result [16].

5.3 Training

Once the network has been initialized and the training input space prepared the network is readyto be trained.

5

International Journal on Cybernetics & Informatics ( IJCI) Vol.2, No.1, February 2013

6.EXPERIMENTALRESULTS

Proposed method was implemented using MATLAB version 7.11 and configuration of personalcomputer includes: windows XP, Intel(R) Pentium4, 1.80 GHz processor with 1GB RAM.Images of car are taken from internet database. Experiments are done on cars of various colours.The below figures are segmented characters of License plate of car image.

7.CONCLUSIONS

This paper introduces a neural network approach for automatic identification of vehicle Licenseplate. Recognition of character are better then Template matching method. If the accuracy of theresults is a critical factor for a vehicle number plate recognition application, then the networkhaving many hidden layers should be used but if training time is a critical factor then the network

6

International Journal on Cybernetics & Informatics ( IJCI) Vol.2, No.1, February 2013

having single hidden layer should be used.The proposed approach of license plate recognition canbe implemented by policeto detect speed violators, parking areas, highways, bridges or tunnels.Also the prototype of the system is going to be integrated and tested as part of the sensor network.Currently there are certain restrictions on parameters like speed of the vehicle,script on thenumber plate, cleanliness of number plate, quality of captured image, skew in the image whichcan be removed by enhancing the algorithms further.

REFERENCES

[1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16][17]

Ahmad and Mohammad,2009,” Efficient Farsi License Plate Recognition”, IEEE.

Q. GAO,et al.,Aug. 2007,”License Plate Recognition Based On Prior Knowledge“, Proc. IEEEInter. Conf. Automation and Logistics Jinan, China, pp.2964-2968.

Ahmad Radmanesh,June 2005,“A Real Time Vehicle’s License Plate Recognition System”,Proceedings ofthe IEEE Conference on Advanced Video and Signal Based Surveillance (AVSS’03)Vol 4, pp.159-167.

Zhu Wei-gang,Aug. 2002,“A study of locating vehicle license plate based on color feature andmathematical morphology,Signal Processing“, 6th International Conferenceon, pp.748-751, Vol.1.Bernard and Galit, Aug. 2002,“Wavelet-Based Monitoring for Disease Outbreaks and Bioterrorism:Methods and Challenges,Signal Processing”, 6th International Conferenceon, Vol.1.Rafael and Richard,2009,“Digital Image Processing”,3rded. Prentice-Hall Inc.Hui Wuand Bing Li,2011,“License Plate Recognition system”, IEEE.

K. Kanayama,et al.,1991,“Development of vehicle-license number recognition system using real-time image processing and its application to travel-time measurement”,Processing’s of IEEEVehicular Technology Conference, pp.789-804.

D.U. Cho and Y.H. Cho,“Implementation of pre-processing independent of environment andrecognition of car number plate using histogram and template matching”,The Journal of the KoreanInstitute of Communication Sciences,23(1)

Shaohong Wu,October 2011,“A Novel Accurately Automatic License Plate Localization Method”ICEES, pp.155-160.

D. S. Kim, and S. I. Chien, 2001,“Automatic car license plateextraction using modified generalizedsymmetry transform and image warping”,Proc. IEEE Int. Symp. On Industrial Electronics, Vol. 3,pp.2022-2027.

Ming-Kan Wu, et al.,2009,“2-Level-Wavelet-Based License Plate Edge Detection”, 5thInternational Conference on Information Assurance and Security, ICIAS, pp.385-388.

Seyed Hamidreza, et al.,2011,“Extraction and Recognition of The Vehicle License Plate for Passingunder outside Environment”, IEEE,European Intelligence and Security Informatics Conference,pp.234-237.

Parul Shah, et al.,2009,“OCR-based Chassis-Number Recognition using Artificial NeuralNetworks”, ICVES 2009.

Feng Yang, and Fan Yang,“Character Recognition using Parallel BP Neural Network”, IEEE.

A. Akoum, et al.,2009,Two Neural Networks for License Number Plates Recognition”,Journal ofTheoretical and Applied Information Technology.

Subhash Tatale, and Akhil Khare,Sept 2011,“real time anpr for vehicle identification using neuralnetwork”k,International Journal of Advances in Engineering& Technology, IJAET ISSN: 2231-1963-262 , Vol. 1, Issue 4, pp.262-268.

Authors

Sneha G. Patel,M.E. [E & C Engineering] Student, Department of Electronics & communication,SardarVallabhbhai Patel Institute of Technology, Vasad, 388 306, Gujarat, India.

7