【作者】 赵庆鹏； 马华东；

【Author】 Zhao Qingpeng1 Ma Huadong1 1 (Beijing Key Laboratory of Intelligent Telecommunications Software and Multimedia, Beijing University of Posts and Telecommunications, Beijing 100876, China)

【机构】 北京邮电大学智能通信软件与多媒体北京市重点实验室；

【摘要】 几何畸变普遍存在于图像采集设备所获取的图像中,如不对其进行适当的矫正,对后续的处理影响较大.本文分析了常见线性和非线性几何畸变产生的原因,对图像中存在的几何畸变类型做出了判定,结合成像模型和数值分析的方法,提出一个自适应的矫正方案把畸变图像恢复到理想无畸变的图像.实验证明该方案能够有效的矫正复杂环境下的常见几何畸变.1更多还原

【Abstract】 In many applications such as two-dimensional barcode recognition, text recognition and vehicle license character recognition, geometric distortions are ubiquitous among images captured by image acquisition equipment. The influence caused by them is crucial to the subsequent processing. To solve this problem, many rectification theories are proposed. However, most of these theories are based on imaging system models, and they can not rectify the general geometric distortions in different environments for they depend on specific imaging systems. Polynomial coordinate transform method works for general distortions, but it can not be used in real-time image processing system for its low time performance. In this paper, after analyzing the reason of both linear and nonlinear geometric distortions, we present a self-adaptive geometric rectification method by combining the method based on imaging system model and the one based on numerical analysis. There are two sections in this method: geometric transformation and gray interpolation. In the geometric transformation section, to eliminate linear distortions and projection distortions, perspective projection matrices are introduced to transform the images with these geometric distortions to temporary images. Then we define the types of geometry distortions in temporary images. For example, for those which only have lens distortions, methods based on imaging system models are introduced to solve it. We compute related parameters such as distortion coefficients to transform these temporary images to images without geometric distortions. And methods based on numerical analysis are used for those which have irregular distortions. In the interpolation section, we compare both the advantages and shortcomings of simple interpolation, bilinear gray interpolation and cubic convolution interpolation. We choose bilinear gray interpolation as the interpolation method from the perspective of both efficiency and accuracy. The experiment based on two-dimensional barcode recognition shows that this practical method in this paper can rectify geometric distortions effectively in most of the complex environments.更多还原