摘要
本文提出了一种扩展的Hough变换,它可以用于曲线检测、多类目标识别和同类目标失真参数估计。由于对原图象及其Hough变换图象两次进行质心计算,使得失真参数——位移量、尺度因子、旋转角能独立分步地计算。本算法同Krishnapuram—Casasent算法(KC法)相比,具有参数搜索精度和目标识别正确率高、运算速度快、占用内存少及抗干扰性强等优点。在对五种飞机模型图形的失真参数搜索和类型识别中,本算法的处理速度比KC算法提高了25倍,所占内存单元减少到1/5。
An extended Hough-Centroid(HC) transform, based on the Hough transform, template matching and moment invariant theory, is proposed in the paper. The HC algorithm can be used in curve detection, multi-class object discrimination and distortion parameter estimation. By calculating centroid positions of a given image and its Hough transform image, respectively, the distortion parameters of the translation, scale and rotation canbe estimated separately. Therefore, computations of parameters are greatly simplified, It is shown that the HC algorithm has better performances in terms of discrimination accuracy, processing speed, memory and estimation accuracy for parameter searching and multi-class(≥2) object discrimination. Experiment results with edge images of five models of aircraft are provided.
关键词
HC算法
计算机视觉
图象处理
computer vision
image processing
Hough Transform
pattern recognition
