一种基于神经网络的畸变图像校正方法

由于摄像机获取的图像存在几何畸变，因此在对图像进行定量分析前，必须校正畸变。针对传统的畸变图像校正方法，其所建立的畸变数学模型，不仅求解畸变参数复杂、计算量大，且存在很大的数值计算误差的问题．提出了一种基于神经网络的畸变图像校正方法。该方法首先运用图像处理技术从一标准模板的畸变图像中提取样本，然后以样本像素坐标作为网络输入来对神经网络进行训练。由于该训练好的神经网络能够实现畸变图像与非畸变图像之间的映射关系，因此能达到校正图像畸变的目的。最后对该校正方法进行了实验，给出并分析了校正实验结果，校正效果令人满意，并已成功地用于焊接机器人视觉系统。

A Distorted Image Correction Method Based on Neural Networks

Images with geometrical distortions, which are taken by cameras, must be corrected before being analyzed. According to the normal distortion correction method for distorted images which obtains distortion coefficients by setting up a distortion model, but as the calculation is complicated and numerical error becomes a big problem, a distortion correction method based on neural networks is put forward in this paper. First of all, the sample coordinates which serve as input parameters of neural networks are extracted from a distorted template image by image processing technique. Then the neural networks are trained by samples. The trained neural networks can learn any distortion relationship between the normal image and the distorted image. Experiments are done by the new method in this paper, and the correcting results are given and analyzed. The experimental results show the neural networks distortion correction technique is satisfactory and it is used in the vision system of the welding robot.