基于Gabor特征的银行卡号识别算法研究

发布时间：2018-08-08 19:54

【摘要】：随着移动支付的逐渐兴起和快速发展,提供移动支付功能的APP迅速增多,这些APP都要求用户进行银行卡的绑定,输入银行卡号成了必不可少的步骤。可是卡号的位数较多,很难全部记住,手动输入易输错,效率差。商家希望使用移动设备的摄像头捕捉银行卡图像,根据银行卡图像识别卡号并直接将卡号输入到移动设备中,这种输入卡号的方法也给使用者带来了方便。卡号为平面字符的银行卡,其卡号字符的颜色基本为黑色,背景花纹对卡号区域没有影响,人眼通过色差来完成对卡号的识别。卡号为凹凸字符的银行卡,其凹凸字符是通过机器压印而成的,人眼主要通过深度和亮度信息来进行卡号的识别。凹凸字符的银行卡因为卡号字符的颜色与背景区域差别不大,导致卡面上的背景花纹很难与卡号字符分离,这就会给卡号的分割和识别带来很大的困难。本文主要是针对卡号为凹凸字符的银行卡,提出一种适合这类银行卡的卡号识别方法。凹凸银行卡图像的预处理首先是对得到的银行卡图像进行倾斜校正,根据银行卡的特点,本文通过Radon变换计算得到银行卡图像的倾斜角,从而实现对银行卡图像的倾斜矫正。然后根据银行卡号在制作时的位置特点,对矫正后的银行卡图像进行卡号区域的粗提取,去除银行卡上与卡号无关的大部分背景区域,粗略分割得到卡号区域。再对粗提取后的图像进行Canny边缘检测,对得到的边缘图像进行投影分析,根据投影像素点的分布规律,精确分割得到卡号区域。凹凸字符与背景颜色相近,很难与背景分开。背景花纹的存在,也会对分割单个卡号字符造成影响,所以本文使用滑动识别方法来进行卡号的识别。凹凸字符的边缘与背景区域的灰度值有一定的不同,二维Gabor滤波器可以提取一定宽度的字符特征,而且可以从不同角度提取字符的纹理特征,这对背景花纹有一定的抗干扰能力。因此本文使用多角度的Gabor滤波器对单个卡号字符进行纹理特征的提取,再通过PCA和LDA二次降维得到具有较好分类效果的字符特征,利用基于特征的模板匹配算法滑动识别银行卡号,最后通过Luhn算法对卡号的识别结果进行验证。
[Abstract]:With the gradual rise and rapid development of mobile payment, the number of APP providing mobile payment is increasing rapidly. These APP require the user to bind the bank card, and input the bank card number is an essential step. But the number of digits, it is difficult to remember all, manual input easy to input error, inefficient. The merchant hopes to use the camera of the mobile device to capture the image of the bank card and to recognize the card number according to the image of the bank card and directly input the card number into the mobile device. This method also brings convenience to the user. The color of the card is black, the background pattern has no effect on the card number area, and the human eye can recognize the card number by color difference. The card is a bank card with concave and convex characters, which is imprinted by machine. The eyes recognize the card numbers by depth and luminance information. The bank card with concave and convex characters has little difference between the color of the card number character and the background area, so the background pattern on the card surface is difficult to separate from the card number character, which will bring great difficulties to the segmentation and recognition of the card number. This paper presents a method of card number recognition for the card with concave and convex characters. The preprocessing of bump bank card image is to correct the bank card image firstly. According to the characteristics of bank card, this paper calculates the tilt angle of bank card image by Radon transform, and realizes the tilt correction of bank card image. Then according to the characteristics of the position of the bank card number, the corrected bank card image is extracted from the card number region, and most background areas which are not related to the card number on the bank card are removed, and the card number area is roughly segmented. Then the coarse extracted image is detected by Canny, and the edge image is projected and analyzed. According to the distribution rule of projection pixels, the card number region can be accurately segmented. Concave and convex characters are similar in color to the background and difficult to separate from the background. The existence of background pattern will also affect the segmentation of single card number characters, so this paper uses sliding recognition method to identify the card number. The edge of concave and convex character is different from the gray value of background area. Two-dimensional Gabor filter can extract character feature of certain width and texture feature of character from different angle. It has certain anti-interference ability to background pattern. In this paper, we use multi-angle Gabor filter to extract the texture feature of single character, and then get the character feature with better classification effect by PCA and LDA quadratic dimension reduction. The feature based template matching algorithm is used to identify the bank card number. Finally, the Luhn algorithm is used to verify the recognition result of the card number.
【学位授予单位】：大连海事大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】