利用全局信息提取靶标特征的方法  

Feature Extraction of Target Based on Global Information

作　　者：赵连军[1,2] 刘恩海[1] 张文明[1] 赵汝进[1] 

机构地区：[1]中国科学院光电技术研究所,四川成都610209 [2]中国科学院大学,北京100190

出　　处：《光学学报》2014年第4期156-161,共6页Acta Optica Sinica

基　　金：国家重点基础研究发展计划(2014CB744200)

摘　　要：为了准确提取图像中目标特征,结合靶标的尺寸和特征信息提出了一种基于全局信息的方法。利用霍夫变换(HT)确定图像中包含靶标的目标区域;在目标区域中提取靶标上不同特征区域的中心;利用提取的中心拟合靶标在图像中所占区域的圆心和半径;完成图像上各区域与靶标上对应区域的匹配。实验证明该方法能够有效、准确的提取图像中靶标的特征,实验室内实验中靶标上特征区域中心提取精度为0.09pixel,实验室外提取精度为0.12pixel。在序列图像处理时,利用前一帧图像的结果可以有效降低计算量,提高提取精度。In order to extract center and radius of target in the image, a method based on global information of target is proposed. Hough Transform （HT） for circle detection is provided as a preprocessing procedure for target detection, results of HT are used as reference for segmenting the region for target parameters determination. Region contains target found by HT is called region of interest （ROI）, and square regions on the target and centroids of square regions are extracted in further step. Centroids of square regions are used to fit circle for determination of precise position of center and radius of target. According to center and radius of target, other regions on the target are detected, and parameters of these regions are deduced. Correspondences between parameters extracted from the image and model target are calculated. Experiments operated in laboratory show that position precision of centroids of regions in ROI is 0.09 pixel, and experiments outside of laboratory show that the precision is 0.12 pixel. Complexity is simplified and position precision is improved in accordance to previous image when sequence images are under processing.

关 键 词：特征提取 目标识别 霍夫变换 曲线拟合 机器视觉 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]