An Efficient Technique for Segmentation ofCharacters of Vehicle Identification by ISERP ISERP

Y.S. Pratyusha et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES Vol No. 5, Issue No. 2, 187 - 194

An Efficient Technique for Segmentation of Characters of Vehicle Identification Number Using Watershed Algorithm Y.S. Pratyusha,

N.S. Murthy

Department of ECE V. R. Siddhartha Engineering College Vijayawada-520007, India

nsmmit@gmail.com,

srk_kalva@yahoo.com

available which are different in terms of contrast and background in. The license plates can be deliberately altered in fraud situations or replaced (e.g., with a stolen plate), which is not possible with the VIN. The technique of identifying VIN is new and more challenging than license plate recognition. This is because license plates have a fixed format (at least) region-wise, while VIN number differs according to the make and the model of the vehicle. Further, the license plates have a standard and a visible look with foreground and background with stark contrast, making character segmentation comparatively easy. However, the VIN embossed on the car body and that placed at the dashboard have similar foreground and background contrasts which makes it difficult to decipher.

Abstract: The license plate detection has become an important application of Intelligent Transportation Systems. This can be used for a broader range of applications like toll collection, reduce congestion and pollution. Since license plates can be replaced, stolen or simply tampered, they are not the ultimate answer for vehicle identification. The objective is to develop a system where by Vehicle Identification Number (VIN) is digitally photographed and then identified by segmenting the characters from the images captured. In this paper a novel approach for segmentation of characters of VIN using watershed algorithm is presented. Initially the VIN was kept in focus and images were captured. The images were then subjected to pre-processing which consists of image processing algorithms. These images were further efficiently processed by considering the distance transforms. The elapsed time and entropy results give an analysis for increasing the efficiency and good performance of watershed segmentation.

Department of ECE V. R. Siddhartha Engineering College Vijayawada-520007, India

sita_pratyusha@yahoo.co.in,

Dr. K. Sri RamaKrishna

Keywords: Vehicle Identification Number; Distance

Transforms; Watershed Algorithm I.

INTRODUCTION

The Vehicle Identification Number (VIN) is a unique identification number for every car manufactured. A registration number can be changed, but the VIN is constant throughout the lifetime of the car. Moreover, the registration number is not unique to a car since it can be carried forward to another car. The Vehicle Identification Number can be found most likely close to the engine, on the body frame, depending on the make or brand of the car. The wall behind the motor (firewall) separates the engine bay and the vehicles dashboard in the cabin. The VIN, in almost all cases, is in the middle, at the top of the firewall and either stamped in the same colour as the paintwork or on silver “credit card” sized plate. The other types of VIN are also

ISSN: 2230-7818

II.

RELATED WORK

In the past decades the telematics industry has increased in size and scope so that technology developed in this field is used to perform tasks ranging from updating accounts on toll roads to tracking vehicle progress for onboard navigational systems. The scope of vehicle telematics seems to be increasing where in VIN provides information on where and how people are driving, what toll roads they are using and what features are adding to their cars. This information could be used as a powerful safety resource. The reason that VIN is popular for telematics purposes is that each car’s VIN is unique and that it can be decoded to give basic information concerning the car. Furthermore it is a permanent identifier and cannot legally be altered by sale or repair. Most cars also have their VIN in bar code form which means it has the potential to be used by remote sensor systems as a tracking identifier. A number of techniques to segment each character after localizing the license plate in the image have been developed, such as feature vector extraction and mathematical morphology [6], and Markov random fields (MRFs) [7]. The work in [6] indicates that the method could be used for character segmentation in plates with indistinguishable characters during off-line operation,

Page 187

Y.S. Pratyusha et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES Vol No. 5, Issue No. 2, 187 - 194

A. Pixel-Based Segmentation Point-based or pixel-based segmentation is conceptually the simplest approach used for segmentation. B. Edge-Based Segmentation Even with perfect illumination, pixel based segmentation results in a bias of the size of segmented objects when the objects show variations in their gray values. Darker objects will become too small, brighter objects too large. The size variations result from the fact that the gray values at the edge of an object change only gradually from the background to the object value. No bias in the size occurs if we take the mean of the object and the background gray values as the threshold. However, this approach is only possible if all objects show the same gray value or if we apply different thresholds for each object. An edge based segmentation approach can be used to avoid a bias in the size of the segmented object without using a complex thresholding scheme. Edge-based segmentation is based on the fact that the position of an edge is given by an extreme of the first-order derivative or a zero crossing in the secondorder derivative.

Watershed transform, which can separate an image into many homogeneous non overlapped closed regions, has been widely applied in image segmentation algorithms. Many watershed algorithms have been proposed [7]–[11]. Vincent and Soille proposed a watershed algorithm using immersion simulations [7]. With sorting before the flooding process and with priority queue, this algorithm is dramatically faster than any former ones. Beucher and Meyer’s algorithm also uses immersion simulations [8], [9]. Two types of algorithms are included: one creates watershed pixels and the other produces a complete tessellation of an image. An ordered queue is used in this algorithm, whose concept is similar to that of Vincent and Soille’s algorithm; however, the minima of the input image need to be detected and labeled first, thus increases the complexity of this algorithm. Dobrin et al.. proposed a fast watershed algorithm named split-and-merge algorithm [10]. It can solve the isolated area problems of the former two algorithms when they are employed to create watershed pixels. Although the results obtained are more correct, it is more complex than the other two algorithms. Moreover, watershed transform for video segmentation is often required to produce a tessellation of an image, where the isolated area problem would not occur. Moga et al.. proposed a watershed algorithm suitable for parallel implementation [11]. With parallel computation, the watershed algorithm can be further accelerated. However, it is also complex and requires a powerful platform, which is impractical for general cases.

known that which pixel belongs to which object. The image is parted into regions and we know the discontinuities as the boundaries between the regions. The different types of segmentations are:

but since the algorithm is computationally complex it cannot be proposed for real-time license plate recognition. The method in [5] was developed for license plate segmentation in video sequences. However, the segmentation results were far from suitable for automatic character recognition. By projection method (PM), characters are segmented according to their height and width ranges after their four boundaries are determined. In essence, template matching method (TMM) is another form of PM, but with a more considered segmentation procession and a more precise boundary division. As for clustering method (CM), the license plate characters are segmented by a clustering analysis algorithm in pattern recognition, which overcomes disconnection of characters, but it has a big computational load and slow processing.

III.

REVIEW OF IMAGE SEGMENTATION

All image processing operations generally aim at a better recognition of objects of interest, i. e., at finding suitable local features that can be distinguished from other objects and from the background. The next step is to check each individual pixel to see whether it belongs to an object of interest or not. This operation is called segmentation and produces a binary image. A pixel has the value one if it belongs to the object; otherwise it is zero. Segmentation is the operation at the threshold between low-level image processing and image analysis. After segmentation, it is

ISSN: 2230-7818

C. Region-based Segmentation These methods focus attention on an important aspect of the segmentation process missed with point-based techniques. There a pixel is classified as an object pixel judging solely on its gray value independently of the context. This meant that isolated points or small areas could be classified as object pixels, disregarding the fact that an important characteristic of an object is its connectivity. If we use not the original image but a feature image for the segmentation process, the features represent not a single pixel but a small neighbourhood, depending on the mask sizes of the operators used. At the edges of the objects, however, where the mask includes pixels from the object and the background, any feature that could be useful cannot be computed. The correct procedure would be to limit the mask size at the edge to points of either the object or the background. But how can this be achieved if we can only distinguish the object and the background after computation of the feature? Obviously, this problem cannot be solved in one step, but only iteratively using a procedure in which feature computation and segmentation are performed alternately. In the first step, the features are computed disregarding any object boundaries. Then a preliminary segmentation is performed and the features are computed again, now using the segmentation results to limit the masks of the neighbourhood operations at the object edges to either the object or the background pixels, depending on the location of the centre pixel. To improve the results, feature computation and segmentation can be repeated until the procedure converges into a stable result.

Page 188

Y.S. Pratyusha et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES Vol No. 5, Issue No. 2, 187 - 194

Definition2. Distance is described as an important geometry parameter that is a boundary direction and which is the relations among the pixels in segmented regions. Suppose ) and ( ) in the that there are two pixels ( image, the distance is defined by three methods below: a. Euclidean Distance: (

)

√(

)

(

(2)

)

b. City Block Distance: (

)

|+|

(3)

c. Chess Distance: (

)

|,|

1.) Region Growing Algorithm and Watershed Algorithm The basic idea of region growing method is a collection of pixels with similar properties to form a region. The steps are as follows:

(a) Find a seed pixel as a starting point for each of needed segmentation. (b) Merge the same or similar property of pixel (Based on a pre-determined growing or similar formula to determine) with the seed pixel around the seed pixel domain into the domain of seed pixel. (c) These new pixels act as a new seed pixel to continue the above process until no more pixels that satisfy the condition can be included. The watershed algorithm is more representative in the application of mathematical morphology theory for image segmentation. Watershed algorithm is a region based segmentation techniques image that uses image morphology. Watershed algorithm is an iterative adaptive threshold algorithm. The idea of watershed algorithm is from geography, it see gradient magnitude image as a topographic map, the gradient magnitude in correspond with altitude, the different gradient in correspond with the peak and basin in valley in the image. It sees every object of image (including background) as a separate part and requested there must have one tag at least in the each object (or seed points). IV.

BASIC DEFINITION

A. Basic definition Definition1. Suppose S is a connected region and B is the boundary of the region S. The region area A is defined as

ISSN: 2230-7818

(4)

B. Watershed Algorithm The idea of watershed is drawn from a topographic analogy. Quite naturally the first algorithm for computing watersheds is found in the field of topography [13]. The introduction of the watershed transformation as a morphological tool is due to Digabel et al [1]. Watershed is then approached theoretically by F. Maisonneuve and used in numerous grayscale segmentation problems. Currently, it is being studied from theoretical, practical, and algorithmic points of view. The watershed transform applied to the image does not produce contours of the features. On the contrary, it partitions the image into the associated areas by the intensity gradient and considers the gradient image as a topographic relief, where the intensity of a pixel denotes the altitude of that pixel. Each pixel in this digital image is assigned a label during the transformation of the catchments basin of a regional minimum. When finished, the resulting network of dams defines the watershed of the image. Compared to the other methods, the watershed has several advantages as follows [2], [3]. − The gaps are handling properly and the placement of boundaries is at the most significant edges. − The resulting boundaries form closed and connected regions. Watershed algorithm (WA) is a widely-used method in image segmentation and image edge detection. Generally, it is used to process the gray gradient image. Its basic thought is that the image is regarded as topology geomorphology in Geodesy and each pixel value in the image is taken as the altitude above sea level. Each local minimum in the image, including its neighbourhood (namely, Influence Zone in Geodesy), is called catchment basin, and the boundary of the catchment basin is called watershed. The principle of WA is shown in Fig. 1. In fact, the Essence of WA is that finds out the local maximum in the segmented region.

The human eyes have adjustability for the brightness, which we can only identified dozens of gray-scale at any point of complex image, but can identify thousands of colors. In many cases, only utilize gray-Level information cannot extract the target from background; we must by means of color information. Accordingly, with the rapidly improvement of computer processing capabilities, the color image processing is being more and more concerned by people. The color image segmentation is also widely used in many multimedia applications, for example; in order to effectively scan large numbers of images and video data in digital libraries, they all need to be compiled directory, sorting and storage, the color and texture are two most important features of information retrieval based on its content in the images and video. Therefore, the color and texture segmentation often used for indexing and management of data; another example of multimedia applications is the dissemination of information in the network .Today, a large number of multimedia data streams sent on the Internet, However, due to the bandwidth limitations; we need to compress the data, and therefore it calls for image and video segmentation.

) A=∑( ) |( (1) Namely, A is the number of pixels in the region S, including B.

D. Color Image Segmentation Algorithm

Page 189

Y.S. Pratyusha et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES Vol No. 5, Issue No. 2, 187 - 194

WA has many kinds of definition forms and corresponding algorithms, and a classic definition is the water immersion algorithm in literature [8]. Suppose that the gray maximum and minimum in the original image P are marked for Max (P) and for Min (P), the coordinates sets of the local minimums in P are respectively, S ( ) is the coordinates set of the region connected with , ) is the gray value of the pixel ( ) and ( ) ( ) is f( ) and f ( )<T, namely: the set of the coordinate ( (

)(

)= { (

)| f (

)<T}

(5)

( )=⋃

*S(

)⋂

(

( ) is the set of the pixels in catchment basins when the gray value is T , then:

Figure1. Illustration of Watershed Algorithm

programming languages can be completed in hours using Lab VIEW. The Lab VIEW programs are called virtual instruments (VIs) because their appearance and operation imitate actual instruments. A VI has two main parts a.)Front panel – It is the interactive user interface of a VI. The front panel can contain knobs; push buttons etc. which are the interactive input controls and graphs, LED’s etc. which are indicators. Controls simulate instrument input devices and supply data to block diagram of the VI. Indicators simulate output devices and display the data that block diagram generates. The front panel objects have corresponding terminal on the block diagram. b.)Block diagram – It is the VI’s source code, constructed in LabVIEW’s graphical programming language, G. The block diagram is the actual executable program. The components of the block diagram are lower level VIs, built in functions, constants and program execution control structures. Wires are used to connect the objects together to indicate the flow of data between them. The data that we enter into the front panel controls enter the block diagram through these terminals and after execution the output data flow to indicator terminals where they exit the block diagram, reenter the front panel and appear in front panel indicators giving us final results.

)(

)}

(6)

( ) is the set of the pixels in watershed, namely: ( )=P- ( )

(7)

The water immersion algorithm is a circulation iteration procession and is described as follows: 1.) Algorithm (Image)

Input: Image is a gray image; Output: ( ) is the set of the pixels in catchment basins, namely, segmentation results, which correspond with the watershed region. Step1. Initialization: S ( ) is originally formed, let ( ) Step2. ( ) ( )is computed according to (5); Step3. ( ) is computed according to (6); ( ) Step4. If T≤ , then go to Step2; Otherwise, iteration procession ends, go to Step5; Step5. W (T) is computed according to (7) and the set of the pixels in corresponding watershed is obtained. ( ) is returned, the algorithm ends. Obviously, in the algorithm procession, Influence Zone is obtained according to previous pixels in the catchment basin, and then the new valley is merged to get present results. V.

ISOLATION OF REGION OF INTEREST

A. Implementation Using Lab VIEW: Lab VIEW or Laboratory Virtual Instrument Engineering Workbench is graphical programming language software used for data acquisition and instrument control. The programs that take a long time using conventional

ISSN: 2230-7818

Page 190

Y.S. Pratyusha et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES Vol No. 5, Issue No. 2, 187 - 194

Test image1

Figure3. Templates of VIN number images

The noise has been removed using filtering techniques of above 3 parts individually .the filter has been applied on each pixel location of fig.2. The filter is more powerful than remaining filters like spatial filters. With the filter, we choose one of two filtering modes: correlation or convolution can be performed. The difference between the two is that convolution rotates the filter by 180 o before performing multiplication. The filter used has been obtained from

VI.

Figure2. Isolated regions using LabVIEW

PRE-PROCESSING

The other images that have been considered for processing are critical and this can be used on various images. The preprocessing of these images is difficult and quite complex. In pre-processing the image, first split the VIN number image (Fig.3) in to 3 parts. Because it includes darker background in last 6 VIN numbers and lighter background in first 6 VIN numbers. The standard VIN number image in fig. 3considered is 138*865.The second image is an example of images which are quite common but it also requires preprocessing as it has noise.

ISSN: 2230-7818

(

)

(

)

VII.

(

∑

) ⁄ (

(5)

)

(6)

∑

BINARIZATION

After pre-processing of image it is converted in to binarized form. Here the binarization is such that the VIN numbers are white and the pixel is 1; background is black and the pixel is 0.If any noise occurs in background image then the morphological operation is done to eliminate the noise. The foundation of morphological processing is in the mathematically rigorous field of set theory; however, this level of sophistication is seldom needed. Most must be the end of a line, and therefore shouldn't be erosion of the thick ridges takes place from the outside. In other words, if a pixel is black, and it is completely surrounded by black pixels, it is to be left alone on this iteration. Running the algorithm both ways shows that it works better. Remember, this is very common in morphological image processing; trial and error is used to find if one technique performs better than another. The pixel must have more than one black neighbor. If it has only one, it is changed in to a continuous line, not a group of interrupted segments. As show in fig.4 is the binarized image.

Page 191

Y.S. Pratyusha et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES Vol No. 5, Issue No. 2, 187 - 194

Figure 5.Euclidean Distance

Figure4. Binarized Image

VIII.

results; In the processing time, City Block Distance has some superiority to Euclidean Distance and approximation to Chess Distance, and it is optimal in entropy and gray results. Therefore in this paper we use City Block Distance to do Distance Transform.

DISTANCE TRANSFORMS

Distance Transform is the procession that turns the binarization image into a gray image. In the gray image, the gray value of each pixel is the least distance between the pixel and the background pixels. Theoretically speaking, if the least distance is obtained above, the global operation must be performed, namely, the least distance is got only after all the distances have been computed between the background pixels and the pixel. If the VIN image size is smaller, the operation is also less. But the image size is big enough, the operation is very considerable. In This paper, our VIN image is considered from fig 3, so we directly compute the distance by three methods in Section 4.1, on which the gray results and the processing time are not the same based. According to Fig. 3, the processing time and entropy are shown in Table 1, and the corresponding effects are shown in Fig 4. The information in the image is more, the more the entropy is, which favors the segmentation of watershed algorithm.

Distance Transforms for test image1

Elapsed Time (sec)

Entropy

S.No

Euclidean Distance

1.065103

9.9501

City Block Distance

0.63148

9.7919

Chess Board Distance

0.717921

9.6068

TABLE1.Processing Time and Entropy for Three Different Distance Transforms

In Table1 and Figs5, 6 and 7 as shown below, Euclidean Distance has no superiority in the processing time, but has some superiority in entropy and gray results; Chess Distance has some superiority in the processing time, but has no superiority to Euclidean Distance in entropy and gray

ISSN: 2230-7818

Figure 6.City Block Distance

Figure 7.Chess Board Distance

The analysis of the performances of these different Distance Transforms algorithms in this section based on the image showed in last section. Euclidean Distance Transform has a higher possibility of â&#x20AC;&#x153;salt and pepperâ&#x20AC;? over segmentation. The reason is that Euclidean propagates to the neighbor pixels in the shape close to a round circle, and it is very easy to form a small island made of a few pixels between different components. When watershed is implemented, the

Page 192

small island will be treated as a separate minimum and forms the â&#x20AC;&#x153;salt and pepperâ&#x20AC;? in the image. City Block Distance Transform has a higher possibility of over segmentation for the components in the image. The reason is that City Block Distance Transform propagates to the neighborhood in the shape of diamond, and it has a very strong tendency to form multiple minima at the centre area of the component where the pixels have different gray levels. When the watershed is implemented, due to the multiple minima at the centre, one component will be segmented into different parts. Chessboard Distance Transform has a better pruning effect due to its square shape propagation. It can effectively remove the jaggedness formed in the Euclidean DT and avoid the components over segmentation caused by City Block Distance Transform. However, Chessboard DT has its own shortcoming. If two components have a high percentage of boundary overlap, Chessboard Distance Transform might be unable to separate the overlapped components. IX.

WATERSHED SEGMENTATION OF CHARACTERS

Figure 8.Watershed and Inverse Watershed for Chess Board Distance

The watershed segmentation gives a clear distinction in terms of an efficient distance transform for segmentation. The distance transforms are mainly used for converting a binary image to gray scale image for watershed algorithm to be performed. The results can be as shown below for different distance transforms.

Y.S. Pratyusha et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES Vol No. 5, Issue No. 2, 187 - 194

Figure 9.Watershed and Inverse Watershed for Euclidean Distance

ISSN: 2230-7818

Page 193

Y.S. Pratyusha et al. / (IJAEST) INTERNATIONAL JOURNAL OF ADVANCED ENGINEERING SCIENCES AND TECHNOLOGIES Vol No. 5, Issue No. 2, 187 - 194

algorithm has been tested on various images and the results shown here are for the image with different illuminations within the same image and the foreground and background isolation was challenging. The algorithm was successfully implemented on various images. The future work includes the recognition of characters and improving the algorithm for recognition. REFERENCES

[1] H. Digabel, C. Lantuejoul, “Iterative algorithms,” In: Proceedings of Quantitative Analysis of Microstructures in Material Science, Biology and Medicine. West Germany: Riederer Verlag, pp. 85–99, 1978. [2] P.R. Hill, C.N. Canagarajah, D.R. Bull, “Image Segmentation Using a Texture Gradient-Based Watershed Transform,” IEEE Trans. Image Processing, pp. 1618–1633, 2003, 12. [3] H.T. Nguyen, M. Worring, R.V.D. Boomgaard, “Watersnakes: EnergyDriven Watershed Segmentation,” IEEE Trans. Pattern Analysis and Machine Intelligence, pp. 330–342, 2003, 25. [4] Y. Cui and Q. Huang, ”Extracting characters of license plates from video sequences,” Mach. Vis. Appl., vol. 10, no. 5/6, pp. 308-320, Apr. 1998. [5]Z. G. Han, S. Y. Lao, Y. X. Xie, et al, “License plates segmentation and adjustment,” Computer Engineering and Applications, vol. 9, pp.210-212, 2003. [6] S. Nomura, K. Yamanaka, O. Katai, H. Kawakami, and T. Shiose, ”A novel adaptive morphological approach for degraded character image segmentation,” Pattern Recognit., vol. 38, no. 11, pp. 1961-1975, Nov. 2005. [7] Y. Cui and Q. Huang, ”Extracting characters of license plates from video sequences,” Mach. Vis. Appl., vol. 10, no. 5/6, pp. 308-320, Apr. 1998. [8] F. Meyer, “Color image segmentation,” in Proc. Int. Conf. Image Processing and Its Applications, 1992, pp. 303–306. [9] E. R. Dougherty, Ed., Mathematical Morphology in Image Processing. New York: Marcel Dekker, 1993, ch. 12, pp. 433–481. [10] B. P. Dobrin, T. Viero, and M. Gabbouj, “Fast watershed algorithms: analysis and extensions,” in SPIE Nonlinear Image Processing V, vol. 2180, 1994, pp. 209–220. [11] A. Moga, B. Cramariuc, and M. Gabbouj, “An efficient watershed segmentation algorithm suitable for parallel implementation,” in Proc. Int. Conf. Image Processing, 1995, pp. 101–104. [12]J. B. T. M. Roerdink and A. Meijster, “The watershed transform: Definitions, algorithmsand parallizations strtagies,” Fundamental informaticae, vol. 41, pp. 187-228, 2000. [13]S.H. Collins, “Terrain Parameters Directly from a Digital Terrain Model,” Canadian Surveyor, pp. 507-518, 1975, 29.

Figure 10.Watershed and Inverse Watershed for City Block Distance

A clear distinction can be observed between the watershed segmentation being applied to different distance transforms from figs 8, 9, 10. X.

CONCLUSION & FUTURE WORK

The proposed Vehicle Identification Number algorithm based on segmentation of characters using watershed algorithm is well suited for intelligent vehicular systems. Initially the image considered was challenging. But the preprocessing algorithm was efficient. Later when the various distance transforms were considered for watershed segmentation it helps in identifying the best suited distance transform for fast performance and increasing the efficiency. The watershed algorithm on the other hand was compared with the inverse of watershed were a distinction can be made. The processor used is Intel Core2Duo with processor of 2.2GHz and the RAM used was 1GB.The MATLAB 7.0 tool has been used for processing. The efficiency of this

ISSN: 2230-7818

Page 194