Robust Real-Time Indian HSRP Recognition System

46th ISTE Annual National Convention & National Conference 2017

International Journal of Advance Research and Innovation (ISSN 2347 – 3258)

Robust Real-Time Indian HSRP Recognition System

Varun Uppal1 , Rajan Kakkar2

1.A.P,A&R Deptt, Gulzar Group of Institutes, Ludhiana (INDIA)

2.A.P,CSE Deptt, Gulzar Group of Institutes, Ludhiana (INDIA)

46th ISTE Annual National Convention & National Conference 2017

International Journal of Advance Research and Innovation (ISSN 2347 – 3258)

Abstract-Robust Indian HSRP (High Security Registration Plate) authentication system is an image processing technique used to recognize and authenticate the HSRPs of the Indian vehicles. This information can be used with or without a database in many real-time applications, such as electronic payment systems (toll payment, parking fee payment), and freeway and arterial monitoring systems for traffic surveillance. This algorithm is composed of nine steps: the image acquisition and preprocessing, registration plate extraction, calculation of the height and width of the registration plate, background color detection, character segmentation, calculation of height and width of each character, IND detection, verification of the registration plate and character recognition.

Index Terms-Automatic license plate recognition (ALPR),Optical Character Recognition(OCR), Template Matching(TM)

Automatic license plate recognition (ALPR) plays an important role in numerous real-life applications, such as automatic toll collection, traffic law enforcement, parking lot access control, and road traffic monitoring[1][2].Registration plate recognition has gained a lot of attention from the research community. Distinctive feature of research work in this area is being restricted to India. This is due to lack of standardization among different registration plates (i.e. the dimensions and the layout of the registration plates. The entire authentication system presented here comprises nine modules. Following is the presentation of every module.

I. IMAGE ACQUISTION AND PREPROCESSING

Image acquisition is the first step in the authentication system .In this proposed algorithm, a high resolution digital camera is used to acquire the image. The input image is then preprocessed before extraction. The preprocessing involves color to gray scale conversion, morphological operation and contrast enhancement. Initially the captured colored image is converted to gray scale image in order to reduce the computation requirements. Subsequently

Morphological operation (i.e. dilation) is performed on the gray scale image in order to improvise the given image by filling holes in the image and sharpen the edges of objects in an image. Next the Histogram Equalization Technique (HET) [3] is used to enhance the contrast of the dilated image. Preprocessing results are shown in Figure 1.

(a) Input color image

(b) Gray scale image

(c) Dilated image

(d) Contrast enhanced image

Fig1.Preprocessing Images

II. IMAGE EXTRACTION

The algorithm employed here is used to extract the

potential registration plate regions. The main principle of identification is the sharp variation in the contrast between the letters and the background

46th ISTE Annual National Convention & National Conference 2017

International Journal of Advance Research and Innovation (ISSN 2347 – 3258)

of the registration plate .There will be frequent changes in the horizontal intensity as numbers and letters are placed in same row(i.e. at identical vertical level).The rows that contain the registration plate will show variation in intensities. First of all, the horizontal and the vertical histograms are passed through a digital low pass filter for smoothing. To remove those areas from the digital filtered image a filter having a particular dynamic threshold[4] is applied. The outputs of such a process are the histograms showing high probability of a containing a registration plate. Now when it is known that which rows and columns are the probable regions of the registration plate, those probable regions are segmented out from the original image and for edge detection Canny operators[5] are used. This is followed by dilation and filling processes. The connected components[6] in the image are then found out from the filled image and each such component is labeled. A bounding rectangle is then created. A bounding rectangle is a rectangle which has a specific weight and height and must cover all connected components. After this the best bounding rectangle is selected by taking into consideration three parameters which are aspect ratio, width of registration plate and total no. of pixels in an ideal Indian HSRP. After cropping the boundary rectangle true number plate is extracted[7].

The final segmented plate is shown in Fig.2.

Fig.2True extracted registration plate.

III.HEIGHT AND WIDTH OF REGISTARTION PLATE

Height is the numbers of rows present in any given image and width is the number of column present in that image. To find out the height and width of the registration plate the differences in the first and the last pixel in vertical and horizontal direction (of the true extracted plate) are used.

IV.BACKGROUND COLOR DETECTION

A series of steps are followed for the calculation of background color detection. The background color of Indian HSRP can be yellow or white or red or blue. Firstly, an RGB value of each pixel of colored registration plate is find out and stored in an array. Secondly, the Standard Deviation (SD) of these RGB values is calculated and compared with the predefined range of respective colors. If the value of

the SD calculated in second step does not match with the predefined ranges of the colors (yellow or white

Table 1 Standard deviation of various colors.

or red or blue), then an error message is shown. The predefined ranges of colors are shown in Table 1.

V.CHARACTER SEGMENTATION

In the first phase of character segmentation firstly the extracted image (Fig.2) is converted into binary image and Canny edge detector is employed to identify those points in an image where brightness changes sharply. Dilation and filling processes are applied thereafter. Again, all the connected component are calculated and out of these only the best connected component is selected by taking consideration only two parameters i.e. the height and the width of the connected components. The characters are then segmented and cropped to edges.Fig.3 shows the results obtained in this stage.

Fig3.Segmented Characters obtained in segmentation stage.

46th ISTE Annual National Convention & National Conference 2017

International Journal of Advance Research and Innovation (ISSN 2347 – 3258)

VI.HEIGHT AND WIDTH OF EACH CHARCTER

The cropped characters obtained in the previous phase provided the basis for the calculation of the height and the width of each character. From the previous cropped images the co-ordinate points of all the four corners of each character is calculated. The differences between the first and the last pixel in y- axis and x-axis are used to calculate the height and the width respectively.

VII.IND DETECTION

By taking into consideration first pixel co-ordinate of the extracted image (see Fig. 2) and the first pixel co- ordinate of the first character an image containing IND section is cropped out for the verification stage. Fig 4 shows the cropped IND image.

Fig. 4 Cropped image of the IND section.

VIII.VERIFICATION

The verification process begins with calculation of the two ratios. These two ratios are the ratio between the calculated heights of the characters to calculated height of the registration plate and the ratio between the calculated widths of the characters to calculated width of the registration plate. If these two ratios fall in a predefined range then the plate is verified else not. Secondly, the height to width ratio of the cropped IND section is computed. If it lies in a certain predefined range then the registration plate is verified else not. Finally, if the background color return error message then the registration plate is not verified. It is due to this phase the authentication of the Indian HSRP is possible. After this phase the character recognition technique is employed to recognize the characters in the Indian HSRP.

IX.CHARACTER RECOGNITION

For the recognition of each character, Optical Character Recognition (OCR)[8] using Template Matching(TM)[9] is employed. In this method the sample of the images are used to recognize similar objects in the source image. The matching process moves the template image to all the possible positions in a larger source image and computes a numerical index that shows how well the template matches the image in that position. Matching is done pixel by pixel. To identify each of the segmented characters of the registration plate through TM the template size of 42X24 is used. Firstly, the segments received from the identification are clipped in such a way that the four corners of new segments coincide with the four corners of the pattern to be identified. The clipped pattern is now resized to sample pattern size i.e. the size of the pattern wit which the correlation is to be found. Thereafter the correlation of this segment and sample character pattern is found. The value of correlation is always between -1 and +1.

Fig.5 Snapshots of MATLAB command window displaying recognized characters.

46th ISTE Annual National Convention & National Conference 2017

International Journal of Advance Research and Innovation (ISSN 2347 – 3258)

The sample pattern that corresponds to maximum correlation is accepted and pattern which was to be identified is labeled as that particular character. Finally, each identified character is written into the text file. The process is repeated until all characters are identified.Fig.5 shows the recognized characters.

X.EXPERIMENTAL RESULTS

required. Also, the issues like strains, blurred images and different font style and sizes are needed to be taken care of. The proposed HSRP detection system is working on single image at a time. So this HSRP detection work can be further extended to work on multiple images at a time and also to minimize the errors as possible.

46th ISTE Annual National Convention & National Conference 2017

International Journal of Advance Research and Innovation (ISSN 2347 – 3258)

In order to test the system proposed above the methods are tested on 124 images (of the Indian HSRPs) collected in different illuminations, under different exposure of light with different angles. In the process of final detection after optimizing the parameters like brightness, contrast and gamma, adjustments, optimum values for lightening and the angle from which the image is taken an overall efficiency of 97% is achieved from this system. Table

2 displays the accuracy obtained in various processes.

Table 2. Accuracy obtained in various processes

It is seen that the proposed algorithm can extract the license plate in normal shapes as well as in skewed shapes. The algorithm also shows its efficiency in recognizing damaged or dirty license plates. The proposed algorithm also shows its robustness in the face of low resolution. It has been observed that the system designed is stronger in the case when the license plate is overexposed or underexposed to light.

XI.CONCLUSION

This paper mainly designs a complete license plate authentication system. It is composed of image preprocessing, the license plate positioning and segmentation, characters segmentation, and character recognition. This paper also designs an extension to ALPR system by making it an authentication system instead of recognition system because it involves verification processes as well. The proposed algorithm is fast enough. Though the system achieves an overall efficiency of 97%., it is required that for the task as sensitive as HSRP detection, tracking stolen vehicles and monitoring vehicles for home land security an accuracy of 100% cannot be compromised with. Therefore to achieve this, further optimization is

REFERENCES

[1]G.Liu, Z.Ma, Z. Du, and C. Wen, “The calculation method of road travel time based on license plate recognition technology,” in Proc. Adv.Inform. Tech. Educ. Commun. Comput. Inform. Sci., vol.

201. 2011, pp.385–389.

[2] Y.-C. Chiou, L. W. Lan, C.-M. Tseng, and C.-C. Fan, “Optimal locations of license plate recognition to enhance the origin-destination matrix estimation,” in Proc. Eastern Asia Soc. Transp. Stu., vol.

8.2011, pp. 1–14.

[3]Vaddi,R.S. ; Vankayalapati, H.D. ; Boggavarapu, L.N.P. ; “Comparative analysis of contrast enhancement techniques between Histogram Equalization and CNN”, Proceedings IEEE International Conference on Digital Object Identifier, Publication Year: 2011, Page(s):106-110 [4] htm

[5]CANN Y J A. Computational approach to edge detection [J]. IEEE Trans Pattern Analysis and Machine Intelli2gence, 1986, 8 (6):6792698.

[6]E. Davies Machine Vision: Theory, Algorithms

and Practicalities, Academic Press, 1990, Chap. 6. [7]R.C.Gonzalez.R.E.Woods and S.L. Eddins. Digital Image Processing. Using MATLAB, Publishing House of Electronics Industry, pp. 280-281,2005.

[8] C. Y. Wang, “Edge detection using tempate

matching,” Image Processing ,

Threshold Logic, Analysis, Filters, Duke University, p. 288, 1985.

[9]Nadira Muda, NikKamariahNik Ismail, SitiAzami Abu Bakar, JasniMohamadZain ,” Optical Character Recognition By Using Template Matching”.