基于Harris角点的内窥镜图像变形全自动校正算法

A Fully Automatic Endoscopic Image Correction Method Using Harris Corner Detection

提出一种基于Harris角点检测的全自动内窥镜图像变形校正算法.对一幅黑白棋格图像进行Harris特征点检测并将特征按所属直线进行自动归类.引入关于全体变形参数的Harris角点子集直线度函数并对它进行Levenberg-Marquardt迭代优化来寻找最优变形参数.获取图像时无需限定相机与标定图案的相对姿态,算法执行中无需用户交互式干预.实验结果表明本文算法的校正结果使得内窥镜成像很好地满足线性相机模型,具有亚像素级精度,是一种快速简单灵活的校正方法,尤其适合临床应用.

We present a fully automatic endoscopic image correction method using Harris corner detection. Harris corner detection on a chessboard pattern image is performed and the interest points are grouped automatically according to lines they belong to. A metric function taking the distortion coefficients as parameters which measures the straightness of the lines consisting of the interest points is introduced. The distortion coefficients are determined by minimizing the metric function using the Levenberg-Marquardt method. No relative pose of the camera with respect to the chessboard pattern is assumed during the image acquisition and no user interaction is needed when the proposed algorithm is running. Experimental results show that our approach yields a good linear camera model with sub-pixel accuracy and is a simple flexible method especially for clinical application.