**Application of Gaussian-Hermite Moments in License**

Lin Wang, Xinggu Pan, ZiZhong Niu and Xiaojuan Ma *Guizhou University for Nationalities China* 

#### **1. Introduction**

12 Will-be-set-by-IN-TECH

84 Recent Advances in Document Recognition and Understanding

Vuori, V. Laaksonen, J. Oja, E. et Kangas, J.(2001). Experiments with adaptation strategies for a

3, pp: 150-159, 2001.

prototype-based recognition system for isolated handwritten characters, *IJDAR*, Vol.

In recent years, many researches on intelligent transportation systems (ITS) have been reported. ITSs are made up of 16 types of technology-based systems divided into intelligent infrastructure systems and intelligent vehicle systems. As one form of ITS technology, vehicle license plate recognition (VLPR) is one of important techniques that can be used for the identification of vehicles all over the world. There are many applications such as entrance admission, security, parking control, airport or harbor cargo control, road traffic control, speed control, toll gate automation and so on. LPR, as a means of vehicle identification, may be further exploited in various ways such as vehicle model identification, under-vehicle surveillance, speed estimation, and intelligent traffic management. Character recognition is an essential and important step in an ALPR system, which influences the overall accuracy and processing speed of the whole system significantly (Jia, 2007 & Christos-Nikolaos et al., 2008).

However, few researches have been done for recognition of car plate character. Neural network method has been employed to recognize car plate characters. The method can achieve promising performance if the quality of the given car plate image is well. However, the quality of image taken for car plates is not always well. This is due to the operating conditions (e.g. dust on the car plates) and distortion or degraded because of poor photographical environment. Experiments have shown that it is difficult to achieve high car plate recognition rates only by extracting features from character are fed into neural network method (Rosenfeld, 1969, Huang et al., 2008).

Moments, such as geometric moments and orthogonal moments, are widely used in pattern recognition, image processing, computer vision and multiresolution analysis (Shen, 1997, 2000; Wu & Shen, 2004; Wang et al., 2004, 2007; ). We present in this paper a study on Gaussian-Hermite moments (GHMs), their calculation, properties, application and so forth.

In this paper, we at first the plate image by preprocessing algorithms (skew corrected, character segmentation, binary image and normalized) before recognition. Then, we propose the GHMs features as the input vector of BP neural network. Our analysis shows orthogonal moment's base functions of different orders having different number of zero crossings and very different shapes, therefore they can better reflect image features based on different modes, which is very interesting for pattern analysis, shape classification, and detection of the moving objects. Moreover, the base functions of GHMs are much more smoothed; are thus less sensitive to noise and avoid the artifacts introduced by window

Application of Gaussian-Hermite Moments in License 87

(a) (b) (c)

Then, calculating for segmentation points by the vertical projection and merging fragments that belong to the same character. The average filter with length s=3 reduce noise. The characters will be extracted from the vertical projection histogram of plate image. The

All the character images are binarized using an anto-adaptive threshold. We propose many iterations algorithm for obtaining the optimal thresholds for segmenting gray scale images. The image background is black (gray value is 0) and the characters are white (gray value is

Given that F(x,y) is the edge image and T is a predefined threshold Tn is a dynamic

 <sup>1</sup> min[ ( , )] max[ ( , )] <sup>2</sup>

 (,) (,) 1 1 <sup>1</sup> (,) (,) <sup>2</sup> *<sup>n</sup> Fxy T Fxy T*

where M is number of pixel with its gray value less than T , N is number of pixel with its

If the intensity of every pixel value is greater than T, the pixel is set to white; otherwise, it is

Characters segmented from different car plates have different sizes. A linear normalization algorithm is applied to the input image to adjust to a uniform size.In our implementation, character blocks are normalized to a fixed size of 32\*16 pixels.Assume the horizontal and vertical projections of the original image F be H and V, respectively. The normalization

*F x y F x y* (1)

*Fxy Fxy M N* (2)

threshold. The following equation is used to obtain a local optimal threshold value.

Fig. 2. Frame removal: (a) Original image; (b) The horizontal cut lines after corrected; (c)

Frame removed

255).

gray value larger than T.

**2.4 Normalization** 

extracted characters are given in Fig. 3.

**2.3 Adaptive threshold for image binarization** 

set to black. The binary image is given in Fig. 3.

**Loop** : if (*T Tn*)

end if

end loop; else if *|T- Tn|* <0.6,

*T=Tn*; goto loop;

position (m,n) of (i,j) is obtained by

function's discontinuity (Fernandez-Garcia, & Medina-Carnicer 2004). Our method can have potential applications in video retrieval, and in other related areas of video information processing.

This paper is organized as follows. First, Section II introduces methods for image preprocessing. Section III presents the orthogonal Gaussian-Hermite moments and their behaviors in the license plate character image. In Section IV, proposes the GHMs features as the input vector of BP neural network for recognizing characters. Section V shows some experiment results. Finally, conclusions are drawn and with the future work discussed as well.

#### **2. Image preprocessing**

#### **2.1 Orientation method for skew correction**

The skew correction of license plate is an important step in ALPR. The license plate inclination is determined by the direction of the boundary. In order to find such direction. In our paper (Ma et al., 2009), the license plate image is firstly divided into a set of 55 nonoverlapping blocks. The local orientation of each block is estimated by gradients [Gx,Gy] of pixels in the block. It may reduce much processing time. Next, the direction histogram which can reveal the overall orientation information in the license plate image is counted. The skew angle of license plate is detected by the local maximum of the direction histogram. This approach can solve the direction detection problem in a very straightforward and robust way under various conditions. Fig. 1 gives some images before and after skew correction.

Fig. 1. Corrected license plates
