Comparative Study of Differet Techiques for Licese Plate Recogitio By Jitedra Sharma 1, Prof Khushboo Saxea 2, Prof Amit Sihal 3 1, 2, 3 Departmet of Iformatio Techology, Techocrats Istitute of Techology, Bhopal jamofperso@gmail.com, kskhusboosaxea26@gmail.com ABSTRACT Licese Plate detectio ad recogitio is a key techique i most of the traffic related applicatios such as searchig of stole vehicles, road traffic moitorig, airport gate moitorig, speed moitorig ad Automatic parkig access cotrol. It is simply the ability to automatically extract ad recogitio of the vehicle licese umber plate s character from a captured image. The Licese plate recogitio systems have established a lot of iterest from the research commuity because there is a certai limitatio i particular territory, city or coutries ad lack of equivalece amog differet licese plates. I This paper, we try to give a ehace ad comprehesive view of the research work made i the area of LP detectio ad recogitio ad the techiques employed i developig a Licese Plate Recogitio (LPR) system. I this aalysis we ot oly study ad also compare the Licese Plate Recogitio Techiques based o differet criteria like plate regio extractio, character segmetatio ad character recogitio rate usig ANN, ad Template Matchig alog with its pros ad cos. Keywords LP detectio, Licese Plate Recogitio, Artificial Neural Network, Radial Basis Fuctio Neural Network, Template Matchig. 1. INTRODUCTION Licese Plate Recogitio (LPR) is a image techology used to idetify plates for their vehicles. This techology is gaiig popularity i security ad traffic facilities. The purpose of LPR was to build a system capable of automatically recordig of the licese plate umbers of passig vehicles travelig dow a roadway. [1] Nowadays, the licese plate recogitio is broadly employed i traffic maagemet to recogize a vehicle whose ower has despoiled traffic laws or to fid stole vehicles. Vehicle Licese plate detectio ad recogitio is a key techique i most of the traffic related applicatios such as searchig of stole vehicles, road traffic moitorig, airport gate moitorig, speed moitorig ad Automatic parkig lots access cotrol. [4] I order to solve such Licese plate remais the priciple vehicle idetifier despite the fact that it ca be deliberated altered i fraud situatio (with stole plate). The Licese plate recogitio system (LPRS) was itroduced i 1976 at the police scietific developmet brach at the Uited Kigdom (UK). It is similar to the Automatic umber plate recogitio (ANPR). However, it has achieved must greater iterest durig the last few decades alog with the upgradig of the digital cameras ad icreased i computatioal capacity [2]. The Purpose of this Automatic licese plate recogitio system is to impact illicit drug traffickig o the regioal stability of the metro city i particular i combiatio with other preset destabilizig factors. It is simply the ability to automatically extract ad recogitio of the vehicle licese umber plate s character from a captured image. It is essetially cosist of frame grabber or camera that has the ability to grab a image, fid the locatio of the umber i the figure ad take out the character of a istrumet of character recogitio to covert the pixel ito easily readable character. It ca also be used i highly sesitive areas like military zoes or area above top govermet head offices. The detectio of the stole vehicle ca be doe i a efficiet maer by usig ANPR system istalled o the highways [2] [3]. The LPRS operates i three steps, the first step is the detectio ad capturig a vehicle image, secod step is the detectio ad extractio of umber plate i a image ad last step is use image segmetatio techique to get the idividual character ad Optical Character Recogitio (OCR) to idetify the idividual character with the help of database stored for each ad every alphaumeric character [4]. The Licese Plate Recogitio System (LPRS) is ormally made up of four parts, where each of the process cotais sub process. A. LP image Locatio Licese Plate Locatio is a itegral ad iseparable part of the system, as it locates the plate that ecloses the licese plate umbers. The whole cocept depeds o the edges of the licese plate oly because there are a huge data i the image. The extractio of multiple licese plates from a image with a complex backgroud is the mai factor. Differet processes performed to extract the licese plate. The extractor gives its output to the segmetatio part. [4] B. Character Segmetatio Licese plate segmetatio, process is also called as the Character Separatio. After the licese plate images are extracted from a picture, ad the fid idividual character i the licese plate to recogize it. I the segmetatio of licese 1
plate characters, licese plate is first coverted ito biary image the characters divided ito segmets to essetial parts obtaiig the characters separately [4]. Licese plate Segmetatio is useful to the licese plate i arrage to outlie the idividual characters. Licese Plate Segmetatio, which is referred to as Character Isolatio takes the regio of iterest ad attempts to split it ito idividual characters. C. Character Recogitio. Licese Plate Recogitio is the last step of the LPR system. This step is the mai part of the recogitio process which decides the accuracy ad recogitio rate of the system. This Recogitio ivolves about to recogize the characters of the licese plate umbers ad character [4] [5]. Before the recogitio the licese plate characters are ormalized. Normalizatio is to improve the characters ito a block cotaiig o added white spaces (pixels) i all the four sides of the characters. I this Stage, the licese plate character images that are take out from the licese plate image have to be recogized. It is actually the process of the character recogitio of the licese plate characters. The character recogitio of the licese plate ca be fid out through Neural Network, Template matchig, Hough Trasform, Radial Basic Fuctio. 2. LITERATURE REVIEW The literature review is divided ito two sectios. I first sectio we describe the Licese Plate Detectio usig RBF Neural Networks ad i secod sectio we describe some more techiques i detail for Licese Plate Detectio usig Template Matchig. A. Review for licese plate detectio usig 1. Vehicle Licese Plate Recogitio Based o Text-lie Costructio ad Multilevel RBF Neural Network I this techique author gave a ew approach for licese plate recogitio usig the text-lie costructio ad multilevel. The locatio of the licese plate will be determied accordig to the text-lie costructio result ad the characteristics of the vehicle licese plate character arragemet. Ad the the locally best adaptive biarizatio is utilized to make more accurate licese plate localizatio. After the licese plate localizatio, the method of vertical projectio iformatio with prior kowledge is proposed to segmet character ad extract the statistic feature. The multilevel classificatio RBF eural etwork is the used to recogize with feature vectors as iput so that; the recogitio result is reliable ad satisfied by the accuracy requiremet of the Itelliget Trasportatio Systems. [2] Use of Multilevel classificatio of is improvig the accuracy of the recogitio of the characters. The system is uable to properly recogize some characters ad umbers like 2, 0, 7 ad more; due to this the recogitio rate is lower compare to the other characters. 2. Automatic Licese Plate Detectio ad Recogitio Usig Radial Basis Fuctio Neural Network I this work authors has used the Radial Basis Fuctio Neural Network. The used for the plate detectio ad recogitio of the licese plate. It also uses the differet image pre-processig processes like edge detectio, image dilatio, filterig, fillig ad smoothig for improvig the quality of the image. The pre-processig firstly performs gray scale coversio of the image. Ad the the gray scale image is coverted ito black ad white image usig sobel edge detector ad morphological method. To fill the gaps, flood fill algorithm is used. After completio of all this steps, the pre-processed images are used for LP detectio. For this it uses threshold value, the objects those satisfy the maximum ad miimum threshold value are the classified as Plate ad No Plate categories. The is used for LP detectio. The objects those are classified as Plate are fially used for traied the. For character detectio the is used with feature extractio process, the origial features are extracted ad feed ito NN for classify the characters. [7] The character segmetatio rate is 99%. The filterig process used for character recogitio from other images, it may leads to erroeous Licese plate detectio. 3. A Algorithm for Licese Plate Recogitio Usig Radial Basis Fuctio Neural Network The authors proposed a ew techique for the licese plate recogitio system. Sobel edge detector, vertical ad horizotal projectio ad mathematical morphology (MM) operatio used for LP regio. The character segmetatio with the help of differet process like Biarizatio based o core regio, Uifyig the backgroud color, coected compoet aalysis, ad Extract character size judgmet. Ad used for the character recogitio process. This uses sigle hidde layer which provides maximal output accordig to the iput. [8] Coected compoet aalysis is used for the character segmetatio. It elimiates the compoet whose height ot match with the characters from the plate ad provides the segmeted characters. Ad with the help of this it fids the exact height of the characters. The system may fail if the texture of the plate is ot clear ad the Licese plate regio is short of the threshold of projectio operatio. 4. A Smart Hybrid Licese Plate Recogitio System Based o Image Processig usig Neural Network ad Image Correlatio The authors ivolved i the field of the licese plate recogitio system as a variety of Licese Plates (LP). It uses the image processig for the character segmetatio. Image correlatio ad Neural Network with LVQ (Learig Vector Quatizatio) used for recogitio process. The hybrid 2
method is better for recogitio, it uses image correlatio ad NN with LVQ with icreases the probability of recogitio of the right character. The character first passed to the image correlatio if its value is greater or equal to the ½ the characters is recogized correctly. If it less tha ½ the character is wrog. Ad fially it go it passes to the NN for recogitio ad accepted the character as the right oe. [9] The character segmetatio process segmet the character 100%, which removes all uwated areas. The mai problem i the system is that if the correlatio coefficiet value is greater or equal tha ½, it accepts the character as the correct whereas character may be wrog. Ad the correct character is gettig by the eural etwork. But it accepts the character recogized by correlatio ot by the eural etwork. B. Review for licese plate detectio usig Template matchig 1. Licese Plate Recogitio System for Idia vehicles The authors ivolved i the field of the licese plate recogitio system as from field of image processig ad machie visio i costructio of a Automatic licese plate recogitio system for Idia vehicles term as (ALPRSIV). It icludes algorithms, which make sure process plate ormalizatio, process of suitable character segmetatio, ormalizatio ad recogitio. All these methods achievig ivariace of systems towards image skew homographs ad a variety of plate coditio. The process of skew correctio which uses the Gamma correctio method, after that the image is coverted to gray scale for itesity adjustmet. Horizotal ad vertical projectio is doe for separate the idividual characters. Remove uwated elemets such as dots ad scratches ad the tries to ehacemet i the segmets. Character recogitio is doe with OCR method. It tries to fid out the earest match. The patter whose correlatio value is earest to the particular character is labeled as the respective character [10] The system gives iformatio regardig the state ad vehicle type accordig to the umber plate after completio all processes. The system has some disadvatages that if the umber plate cotais extra desigs or the fot of the umber plate alters regularly, it creates a problem. The vehicles that do ot follow the stadard rules of Idia Number plates pose a problem to the system. There are umerous umber plates i the vehicle whose coditio is so bad that eve a huma eye caot recogize it. I those cases the system is obviously uable to obtai the expected results. 2. Vehicle-Licese-Plate Recogitio Based o Neural Networks The authors gave the iovative approach for Licese plate recogitio based o Neural Network. To recogize the umber plate the eural etwork chip is used. The chip composed of two modules video image processig module with eural etwork module usig equalized image processig algorithm ad etwork classificatio algorithm. The chip icludes image sesor, CogiMem chip, iterface circuit ad PC moitorig module. The video from the image sesor is set to video module of the CogiMem chip. The feature vector is automatically geerated by the chip from the regio of iterest give by user. The recogitio is show o the PC. [11] High speed of recogitio. The recogitio time of CM1K system was 101µs, while most of the existig techologies eed millisecod processig time. High stability. The CM1K system performs better i stability tha software system, because most of the circuit was itegrated i the FPGA, ad the bottom hardware fiished much amout work of recogitio. O the cotrary, the software system is completely depedet o the PC, so ay delay for the PC might make it loss some plates. Recogitio rate is less. 3. A licese plate recogitio algorithm for Itelliget Trasportatio System applicatios I this techique the author gave a ew approach for image segmetatio. The licese plate recogitio system as uses a algorithm ovel adaptive image segmetatio techique (Slidig Cocetric Widows-SCW) ad coected compoet aalysis i cojuctio with a character recogitio Neural Network. The algorithm was tested with particular atural scee gray level vehicle images of differet backgrouds ad ambiet illumiatio. The camera targeted withi the plate, while the agle of sight ad the distace from the vehicle varied accordig to the coditio if eeded. The character recogitio process is doe by PNN (Probabilistic Neural Network). The SCW is used for the describig the local irregularity i the image usig image statistics. It creates two widows for the pixel of the image. If the ratio of statistical measuremet i two widows more tha threshold set by the user, the the cetral pixel of widows is cosidered as the Regio of Iterest. It gives optimal result, if the ratio of cocetric widow is ear to the ratio of the object is defied.[12] The ovel adaptive techique for segmetatio i.e. Slidig Cocetric widows used for the fidig the regio of Iterest ad for segmetatio the coected compoet aalysis is used. The OCR system is a two layer Probabilistic Neural Network (PNN) with topology, whose performace for etire plate recogitio reached has bee improved easily. The major drawback i the proposed algorithmic sequece revolve aroud the varyig light levels ecoutered durig a 24 hour period ad the effect those lightig chages have o the image beig forwarded to the OCR program as well as due to the physical appearace of the plates. 4. Automatic Licese Plate Recogitio 3
The authors describe automatic licese plate recogitio processes. The LPR system cosists of four steps Plate Localizatio, Preprocessig, Segmetatio ad Normalizatio ad Optical Character Recogitio (OCR). Before applyig the Morphological operator the RGB image is coverted to the gray image ad biary image. Otsu method applied to covert the gray image ito Biary image. The the morphological operator applied o the image to idetify the plate locatio. The the plate regio is preprocessed by applyig the histogram equalizatio techique. The smearig ad morphological algorithms are used to segmet the characters ad segmeted result is ormalized ad fed to the OCR part for recogitio [13] Differet methods are useful for providig better Some of the characters give the erroeous result for cofusig characters like 2 ad Z which decreases the 5. New Morphology-Based Method for Robust Iraia Car Plate Detectio ad Recogitio The author has preseted a ew real time ad robust method of licese plate detectio ad recogitio based o the morphology ad template matchig. Isolatio of the image is the mai phase of the system, from digital image clicked by camera i differet circumstaces. Firstly the image is preprocessed for further recogitio, after that morphological operator is applied for image locatio. The morphological operator is based o the shapes. Morphological process is used for character segmetatio. It removes all small coected elemets. The dilatio operator is applied to separate the character from each other. Ad partitio scaig is doe for character segmetatio. Character recogitio is doe with the help of template matchig process. For this image correlatio method is used. [14] It gives better performace for licese plate localizatio. It correctly fids the locatio of plate up to 97.3%. If the image quality is ot good the system get failed to recogize the characters. 6. Automatic Vehicle Idetificatio by Plate Recogitio The author presets a ew smart ad simple algorithm is preseted for the licese plate recogitio. The LPR works i differet parts: plate regio extractio, character segmetatio ad recogitio. The recogitio process is doe with the help of template matchig. The plate is firstly coverted ito biarized image the the smearig algorithm is applied to fid out the plate regio. After this the morphological operator is applied to get the locatio. The segmetatio is doe with the morphological fuctio. The dilatio operator is used for separate the characters. Before the recogitio, the characters are ormalized. Normaliztio is doe for refie characters ito a block cotaiig o white spaces ad the characters are fit ito equal size for template matchig. Cross correlatio method is used for character recogitio. [15] The detectio of the plate is achieved with the higher accuracy. It shows accuracy for plate extractio 97.6%, segmetatio 96% ad recogitio of characters IS 98.8%. The similarity betwee some characters decreases the 3. CONCLUSION The process of vehicle umber plate recogitio requires a very high degree of accuracy whe the system is deal with differet area of licese plate detectio like automatic parkig, automated toll booth, border cotrol, law eforcemet ad may more. I this paper we have preseted a survey of Licese Plate Recogitio Techiques. It also describes how these techiques are helpful for recogize umber plate ad how to improve recogitio rate, character segmetatio ad plate regio extractio rate. The performace, cocepts alog with pros ad cos of Licese Plate Recogitio techiques are summarized i this paper. Performace aalysis of various Licese Plate Recogitio techiques is discussed i the table below Sr. o Techiques Employed 1 LPR based o Textlie Costructi o ad Multilevel 2 Automatic LPR ad Detectio Usig 3 LPR usig 4 LPR with Image Correlatio Recogiti o Rate (%) Character Segmetati o (%) Plate Regio Extracti o (%) 93 - - 91 99 95 94.3 97 99.3 96.64 100 80 4
ad NN 5 LPR for Idia Vehicles 6 Vehicle LPR i Neural Networks 7 A LPR Algorithm for ITS applicatio s 8 Automatic LPR 9 New 1 0 Morpholog y-based Method for Robust Iraia Car Plate Detectio ad Recogitio Automatic Vehicle Idetificati o by Plate Recogitio 90 - - 91.2 - - 93 99.3 - - - - 92 94 97.3 98.8 96 97.6 Ref: LPR= Licese Plate recogitio; = Radial Basis Fuctio Neural Network; NN= Neural Network; ITS= Itelliget Trasportatio System 4. REFERENCES [1] Hsie-Chu WU, Chwei-Shyog TSAI, ad Chig-Hao LAI A Licese Plate Recogitio System i E- Govermet Iformatio & Security. A Iteratioal Joural, Vol. 15, No. 2. pp. 199-210 2004. [2] Baomig Sha Vehicle Licese Plate Recogitio Based o Text-lie Costructio ad Multilevel RBF Neural Network Joural of computer sciece Vol. No. 6 pp. 246-253 February 2011. [3] Cheta Sharma ad Amadeep Kaur Idia Vehicle Licese Plate Extractio ad Segmetatio Iteratioal Joural of Computer Sciece ad Commuicatio Vol. No. 2, pp. 593-599, July-December 2011. [4] Serka Ozbay ad Ergu Ercelebi Automatic Vehicle Idetificatio by Plate Recogitio World Academy of Sciece, Egieerig ad Techology, pp. 222-225 2005. [5] Auja p. Nagare Licese Plate Character Recogitio System usig Neural Network Iteratioal Joural of Computer Applicatio, Vol. 25, No. 10, July 2011. [6] A. Akoum, B. Daya, P. Chauvet Two Neural Networks for Licese Number Plate Recogitio Joural of Theoretical ad Applied Iformatio Techology 2005 2009. [7] Nureddi A. Abulgasem, Dzulkili Mohamad, Siti Zaito Mohamad Hashim Automatic Licese Plate Detectio ad Recogitio Usig Radial Basis Fuctio Neural Network Idia Joural of Computer Visio ad Applicatios Vol. 1.No. 1 pp. 15-23 2011. [8] Bo Li, Zhi-yuag, Jia-zhog Zhou ad Hua-li Dog A Algorithm for Licese Plate Recogitio Usig Radial Basis Fuctio Neural Network Iteratioal Symposium o Computer Sciece ad Computatioal Techology, pp. 569-572 2008. [9] K. Yilmaz A Smart Hybrid Licese Plate Recogitio System Based o Image Processig usig Neural Network ad Image Correlatio, pp. 148-153 IEEE 2011. [10] Shishir Kumar, Shashak Agarwal & Kumar Saurabh Licese Plate Recogitio System for Idia vehicles Iteratioal Joural of Iformatio Techology ad Kowledge Maagemet, Vol. 1, No. 2, pp. 311-325, July-December 2008. [11] Yi Qig Liu ad Dog Wei, Nig Zhag ad Mi Zhe Zhao Vehicle-Licese-Plate Recogitio Based o Neural Networks, Proceedigs of IEEE Iteratioal Coferece o Iformatio ad Automatio Shezhe, pp363-366, Chia Jue 2011. [12] C.N. Aagostopoulos, I. Aagostopoulos, V. Loumos, ad E. Kayafas A Licese Plate Recogitio Algorithm for Itelliget Trasportatio System Applicatios T ITS-05-08-0094 August 29, 2005. [13] D. Reuka Devi & D. KaagapushpavaUi Automatic Licese Plate Recogitio, pp. 75-78 IEEE 2011. [14] S. Hamidreza Kasaei, S. Mohammadreza Kasaei, S. Alireza Kasaei, New Morphology-Based Method for Robust Iraia Car Plate Detectio ad Recogitio, Iteratioal Joural of Computer Theory ad Egieerig, Vol. No. 2, pp 268-268, April 2010. [15] Serka Ozbay, ad Ergu Ercelebi Automatic Vehicle Idetificatio by Plate Recogitio, World Academy of Sciece, Egieerig ad Techology, pp 778-781, Sep 2007. 5