计算机与现代化

• 模式识别 • 上一篇    下一篇

基于KNN技术的校内网验证码识别

  

  1. (1.武汉邮电科学研究院,湖北武汉430074;2.烽火通信科技股份有限公司,江苏南京210019)
  • 收稿日期:2016-07-06 出版日期:2017-03-09 发布日期:2017-03-20
  • 作者简介:汪洋(1978-),男,江苏南京人,武汉邮电科学研究院、烽火通信科技股份有限公司高级工程师,硕士,研究方向:大数据分析,网络行为学; 许映秋(1990-),男,江苏如皋人,硕士研究生,研究方向:数据分析,多媒体图像处理; 彭艳兵(1974-),男,高级工程师,博士,研究方向:海量数据分析,网络行为分析。

KNN-based Verification Code Recognition Technology on Campus Network

  1. (1.Wuhan Research Institute of Posts and Telecommunications, Wuhan 430074, China;

    2.FiberHome Communications Science & Technology Development Co. Ltd., Nanjing 210019, China)
  • Received:2016-07-06 Online:2017-03-09 Published:2017-03-20

摘要:

随着科技日新月异的发展,验证码技术在网络防护和信息安全方面有着广泛的应用。由于网络攻击手段的提升,验证码技术也在改进。本文采用的校内网验证码是当前网络中最普遍的字符验证码类型,它多元化的背景噪音和字符扭曲粘连的特点,使得验证码很难实现程序自动识别。针对这些特点,本文在背景去噪阶段,提出RGB三原色去噪法;在单个字符切割阶段,采用轮廓差投影法与水滴算法相结合的分割方法。最后得到所有字符模型,再利用KNN算法,进行字符识别,从而得到识别结果。实验结果表明,该方法对有背景噪声和字符扭曲粘连的验证码有很好的识别效果。

关键词: 验证码, RGB三原色背景去噪, 上下轮廓差投影法, 水滴算法, KNN

Abstract:

 With the rapid development of technology, verification code technology in the network protection and information security has a wide range of applications. Since the means of network attacks are upgraded, verification code technology is also continuously improved. The campus network verification code used in this paper is the most common type of character verification codes of the current network. It is of the characteristics of diversified background noise and character distortion adhesion, so the code is difficult to achieve automatic recognition program. For these characteristics, this paper proposes a RGB de-noising method in the background de-noising stage, and adopts a segmentation method which combines contour difference projection method with the water droplets algorithm in the single character cutting stage. Finally, all the character models are get, KNN algorithm is used for character recognition so as to receive recognition results. Experiments show that the method is of good recognition effects for the verification code with background noise and distortion adhesion character.

Key words: verification code, RGB color background denoising, difference between the upper and lower contour projection, water droplets algorithm, KNN

中图分类号: