Car Plate Segmentation Based on Morphological and by idesajith

VIEWS: 113 PAGES: 4

									Short Paper
                       Proc. of Int. Conf. on Advances in Computing, Control, and Telecommunication Technologies 2011



  Car Plate Segmentation Based on Morphological and
                 Labeling Approach
                          Seyed Mojtaba Mousavi, Seyed Omid Shahdi, and S.A.R. Abu-Bakar
                                 Computer Vision, Video and Image Processing Research Lab
                                Faculty of Electrical Engineering, Universiti Teknologi Malaysia
                                                 81310, Skudai, Johor, Malaysia
                                  {mosavi, shahdi}@fkegraduate.utm.my, syed@fke.utm.my


Abstract—Automatic car plate extraction is one of the                     There are some drawbacks with the previous popular methods
interesting research areas in surveillance applications such              to locate the car license plate. One of these method is Template
as traffic control, parking lot and detection of stolen vehicles.         matching. The problem in methods which utilize Template
This topic is divided in two main parts, firstly license plate
                                                                          matching [4, 5] is its requirement for high processing time
localization, and secondly, license plate characters recognition.
In this paper merely the first part is taken into consideration
                                                                          and also its variation due to the scaling. The other common
by means of morphological operations and labeling algorithm.              used method in this area is Hough Transform [5, 6]. However,
Moreover, performance of proposed method is investigated to               by using this method, the clear license plate location could
unveil the ability in extracting of the license plate. Space              not be provided since it does not remove all non-plate regions.
between characters and also height and width of the license               Besides, Hough transform is not reliable for detecting vertical
plate are fundamental issues in morphological operations                  lines due the fact that these lines are more sensitive to noise.
along with dimension of characters in labeling algorithm.                 The Brazilian license plates were localized by the proposed
Results for this method are promising even in tough                       method in [7]. This method was based on detecting the
conditions.
                                                                          vertical edges of grayscale input image. Besides, combination
Index Terms—Sobel edge detection, morphological operation,
                                                                          of region growing method and fuzzy logic based thresholding
labeling                                                                  algorithm is employed in this method to segment the
                                                                          characters.
                        I. INTRODUCTION                                       Ozbay et al. [8] proposed the method to recognize Turkish
                                                                          license plate. Their method mainly consists of three parts
    Nowadays, the fast pace of modern life style has changed              including plate region extraction, characters segmentation
many surrounding products and systems from manually to                    and character recognition. They applied edge detection to
automatic. In this regard, speed of operation process and                 extract the plate region. Then, smearing, filtering and
accuracy are important. To achieve these goals, there are                 morphological algorithms are utilized for segmenting the
many methods which help human. One of most useful device                  characters. Finally template matching algorithm is used for
is computer vision system. Image processing gives us the                  character recognition.
opportunities that use a computer instead of a human as a                     The utilized method in [9] for both license plate extraction
supervisor [1]. In field of automation, image processing also             and character recognition is based on 4 layers back
plays a very substantial role in the car and traffic control              propagation neural network. This paper used license plate
problems. For instance, image processing techniques could                 regulation such as shape, color and characters pattern for
be useful for recognizing and identifying the car plate number            extracting the plate of this region. And in these papers [10-
and applying for some time consuming and hard manually                    17] different character recognition algorithms are used
tasks such as the scenario in multi storey parking lot ticket at          without utilizing license plate localization. In this paper, the
high traffic flow [2].                                                    mixtures of two methods are utilized. One of them is to apply
    Popularity and usefulness of automatic car plate number               the labeling procedure while the other one employs
recognition motivate many researchers to improve and                      morphological method to accomplish this task. The remainder
develop many methods to increase the accuracy and also                    of this paper is arranged as follows. The overview of proposed
decrease the processing time. The whole process can be                    system for plate localization is given in section 2. In section
divided into two main parts, first car plate number extraction            3, basic concepts of labeling and morphological operations
and then recognition of characters included in car plate and              are introduced. The experimental results are presented in
identifying the license. Aim of this paper is confined to extract         section 4. Finally in the last section, conclusion is provided.
the car plate characters.
    Although too many methods are implemented in this area                                   II. PROPOSED METHODS
of research, most of them are applicable for some certain
types of car plate characters. For example, in [3] to extract the             In this paper, labeling and morphological operations are
car plate characters, it is considered the plate’s color is yellow        utilized in order to segment the car license plate. Composition
and the color extraction process is done by searching for                 of these two methods is important since it increases the
only yellow color. Nonetheless, input image including yellow              accuracy when different sort of car plate characters are
color objects can degrade the accuracy results drastically.               penetrated into the system. This fact is due to the elimination
                                                                     84
© 2011 ACEEE
DOI: 02.ACT.2011.03.26
Short Paper
                       Proc. of Int. Conf. on Advances in Computing, Control, and Telecommunication Technologies 2011


of unwanted area by using labeling beforehand of                           As it could be perceived from Fig. 4 (a), making use of solely
morphological operation. The whole process is depicted in                  labeling is not accurate enough to localize the place of license
Fig. 1.                                                                    plate. Because after labeling operation, some comparable
                                                                           regions along with the characters in license plate area are still
            III. CAR LICENSE PLATE SEGMENTATION                            remained. Hence, morphological operation is necessary to
                                                                           fulfill plate localization successfully.
    In this section, labeling and morphological are discussed
individually in details. Nonetheless, beforehand of these                  B. Morphological Based Method
operations the pre-processing step should be taken into                          The morphological operations duty is to apply on the
account to alleviate the affects of noises and intensive                   labeled image. In labeled image the morphological procedures
illumination variations. For this purpose, Gaussian filter is              are employed to fill the gaps between characters for making
utilized which is experimentally outperforms other filters.                the rectangular regions. The morphological operations
A. Labeling Operation                                                      consist of three main sets, namely opening, closing and
                                                                           dilating. The input of this part is emerged from the output of
     Labelling of the connected component is one of the
                                                                           labeling phase. By using sequence of morphological
fundamental operations in many intelligent vision systems. This
                                                                           operations, the region of interest (ROI) is kept and the rests
operation works on binary images by allocating an individual
                                                                           are eliminated. These operations consist of following steps
values to pixels that belong to the same connected area. Two
                                                                           to attain the segmented car plate area or ROI:
widespread types of labelling are 4-connected and 8-connected
                                                                              In the first step, line opening is employed since the
neighbours. In Fig. 2 these two topologies are depicted. This
                                                                                  labeling output, contains many horizontal lines that have
operation is based on labeling and then classifying of labeled
                                                                                  relatively smaller thickness in comparison with height of
objects with relation to their sizes. The original image is
                                                                                  characters in car plate as depicted in Fig. 4(a). These
necessary to convert into the gray-scale image in case of
                                                                                  undesirable lines could be removed by performing this
color image (Fig. 3 (a)). Then the segmented image can be
                                                                                  step as the result is shown in Fig. 4(b).
achieved by means of automatic global thresholding and
                                                                           In the second step, rectangle closing is applied to retain
binarizing (Fig. 3 (b)). It should be noticed that in our database,
                                                                                  the characters which are closed to each other, similar to
all the license plate characters have a white color and this point
                                                                                  the condition within the characters in the car plate license
make the thresholding procedure easier. The labeling results are
                                                                                  as its procedure is demonstrated in Fig. 5(a).
depicted in Fig. 4 (a).
                                                                           By iterating above steps, the only region which is kept
                                                                                  is the car plate region in such a way that the undesirable
                                                                                  regions become smaller and in contrast the plate region
                                                                                  becomes bigger eventually. The illustrative procedure is
                                                                                  shown in Fig. 5(b).




                                                                                 Figure 3. (a) Original gray car image, (b) Binarized image




                                                                                     Figure 4. (a) Labeled image, (b) Vertical opening

             Figure 1. Overview of proposed method                          At the next step, the dilation is carried out in order to
                                                                           reconstruct the car plate region into its primitive size. For
                                                                           further clarification on this issue, refer to Fig. 6(a).
                                                                            At last stage, it is necessary to employ another step of
                                                                           labeling for trimming and further clarifying the place of car
                                                                           plate and removing analogous area which are not the car
                                                                           plate as shown in Fig. 6 (b).
Figure 2. Different connected neighbors: (a) 4-connectedneighbor.              Finally the plate characters can be detected by subtracting
                         (b) 8-connected                                   the last result of morphological procedure from binarized
                                                                           image. The subtracted image is presented in Fig. 7.
                                                                      85
© 2011 ACEEE
DOI: 02.ACT.2011.03.26
Short Paper
                        Proc. of Int. Conf. on Advances in Computing, Control, and Telecommunication Technologies 2011


                    IV. EXPERIMENTAL RESULT                                  severe conditions which are included in our database. Some
                                                                             of the results with these conditions are shown in Fig. 8.
   The experiments are conducted by using Intel core 2 Duo
CPU with 4GB RAM on board. From software point of view,
                                                                                                        CONCLUSION
we utilize MATLAB programming environment version
R2009b. In this section, the database and details on                            In this work, we cope with the challenges in segmentation
experimental results are given.                                              of car plate license in variety of conditions. Both
                                                                             morphological and labeling are utilized in order to come up
A. Database
                                                                             with this task. The outstanding benefit for morphological
    The database used here as an experimental tool contains                  and labeling operations is their ability to segment car plate
cars with the Malaysian license plates. It includes 290                      much faster in comparison with systems which employ neural
numbers of images which are taken in uncontrolled conditions                 network or other complex algorithms. Our results proof the
such as in different illuminations, variable location of car                 robustness of our method against real world conditions by
plate in an image, different sizes of characters within a license            providing satisfactory results in almost whole of the
plate, both single row and double row license plate and some                 employed database. It is further desired to select the
of occlusions besides car plates. These characteristics make                 thresholds using an effective genetic algorithm which is our
our task even more challenging. All the initial images have                  plan for future work.
the standard size 384×288 pixels. Some samples from this
database are depicted in Fig. 8.                                                                    ACKNOWLEDGMENT
B. Experimental Results                                                         This work was supported by Universiti Teknologi
    After penetrating image into the system, labeling and                    Malaysia (UTM) and Ministry of Higher Education (MoHE)
morphology operations are applied as thoroughly described                    under Research University Grant Q.J13000.7123.00H53
in section 3. However, it is necessary to employ another step
of labeling for trimming and further clarifying the place of car                                        REFERENCES
plate and removing analogous area which are not the car
                                                                             [1] C. X. Feng, P. B. Chang, and Z. S. Lin, “A license plate
plate as some instances are shown in Fig. 6.                                 localization method based on region narrowing,” in Machine Learning
    Our system is employed on the aforementioned database                    and Cybernetics, 2008.
to investigate the robustness of our proposed method. The                    [2] G. Ioannis, A. C. Nikolaos, L. Vassili, and K. Eleftherios,
overall accuracy result is 95.5%, in other words, 277 out of                 “Operator context scanning to support high segmentation rates for
290 cars license plates are successfully segmented regarding                 real time license plate recognition,” Pattern Recogn., 2010. 43(11):
                                                                             p. 3866-3878.
                                                                             [3] M. T. Qadri and M. Asif, “Automatic Number Plate
                                                                             Recognition System for Vehicle Identification Using Optical
                                                                             Character Recognition,” in Education Technology and Computer,
                                                                             ICETC, 2009.
                                                                             [4] H. Y. Ping, L. S. Yong, and C. W. Po. A template-based model
                                                                             for license plate recognition. in Networking, Sensing and Control,
                                                                             2004 IEEE International Conference.
 Figure 5. (a) Rectangular closing, (b) Iterating opening and closing
                                                                             [5] P. V. Suryanarayana, S. K. Mitra, A. Banerjee, and A. K Roy,
                                                                             “Morphology Based Approach for Car License Plate Extraction,”
                                                                             IEEE Indicon 2005 Conference, Chennai, India.
                                                                             [6] Y. Yanamura, et al. “Extraction and tracking of the license
                                                                             plate using Hough transform and voted block matching,” in Intelligent
                                                                             Vehicles Symposium, Proceedings, IEEE, 2003.
                                                                             [7] C. A. B. Mello and D. C. Costa, “A Complete System for
                                                                             Vehicle License Plate Recognition”, in Systems, Signals and Image
      Figure 6. (a) Dilated image, (b) Localized license plate               Processing, 16th International Conference on 2009.
                                                                             [8] S. Ozbay and E. Ercelebi, “Automatic Vehicle Identification By
                                                                             Plate Recognition,” World Academy of Science, Engineering and
                                                                             Technology, 2005. 9: p. 222-225.
                                                                             [9] T. Sirithinaphong and K. Chamnongthai, “The recognition of
                                                                             car license plate for automatic parking system,” in Signal Processing
                                                                             and Its Applications, ISSPA, Proceedings of the Fifth International
                                                                             Symposium on 1999.
                                                                             [10] S. Baoming, “License Plate Character Segmentation and
                                                                             Recognition Based on RBF Neural Network,” in Education
                  Figure 7. Extracted license plate                          Technology and Computer Science (ETCS), 2010 Second
                                                                             International Workshop.
                                                                             [11] G. Yang, “License plate character recognition based on wavelet
                                                                             kernel LS-SVM,” in Computer Research and Development
                                                                             (ICCRD), 3rd International Conference on 2011.
                                                                        86
© 2011 ACEEE
DOI: 02.ACT.2011.03. 26
Short Paper
                        Proc. of Int. Conf. on Advances in Computing, Control, and Telecommunication Technologies 2011




                                                  Figure 8. Successfully segmented car plate

[12] F. Jianlan, L. Yuping, and M. Chen, “The research of vehicle            [15] L. Zheng, G. Li, and J. Sha, “The recognition of character in
license plate character recognition method based on artificial neural        license-plate image,” in Computer-Aided Industrial Design and
network,” in Informatics in Control, Automation and Robotics                 Conceptual Design, CAIDC, 7th International Conference on 2006.
(CAR), 2nd International Asia Conference on 2010.                            [16] L. Bo, F. Bin, and L. D. Hui, “Character recognition of license
[13] Z. Ying, X. Yingtao, and D. Gejian, “License plate character            plate image based on multiple classifiers,” in Wavelet Analysis and
recognition algorithm based on filled function method training by            Pattern Recognition, ICWAPR, 2009.
neural network,” in Control and Decision Conference, CCDC,                   [17] R. Huang, H. Tawfik, and A. K. Nagar, “Licence Plate Character
Chinese, 2008.                                                               Recognition Based on Support Vector Machines with Clonal
[14] X. Lu, X. Ling, and W. Huang, “Vehicle license plate character          Selection and Fish Swarm Algorithms,” in Computer Modelling
recognition,” in Neural Networks and Signal Processing,                      and Simulation, UKSIM, 11th International Conference on 2009.
Proceedings of the 2003 International Conference.




                                                                        87
© 2011 ACEEE
DOI: 02.ACT.2011.03.26

								
To top