This paper proposes an enhanced Edge Matching Rate (EMR) to gain good image regis-tration based on Generalized Acreage (GA). Traditional EMR considers only matched pixels sum without concerns of the cause of unmatched...This paper proposes an enhanced Edge Matching Rate (EMR) to gain good image regis-tration based on Generalized Acreage (GA). Traditional EMR considers only matched pixels sum without concerns of the cause of unmatched pixels and the relationship between matched pixels. The modified EMR introduces the new concept of generalized acreage to measure the overlaying parts between the target image and the model. It also defines similarity of local occlusion and of local dithering to measure interference degree. Not only edge points are considered but also non-edge points, occlusion, and dithering. Using the same preprocessing, the experiments match images based on tra-ditional EMR and the proposed EMR separately. Based on the proposed EMR the paper achieves more stable registration and higher precision.展开更多
针对传统图像识别算法匹配正确率低、运行时间较长等问题,文中提出了基于改进ORB-FLANN(Oriented FAST and Rotated BRIEF-Fast Library for Approximate Nearest Neighbors)的工件图像识别方法。对ORB算法特征描述、图像特征匹配算法...针对传统图像识别算法匹配正确率低、运行时间较长等问题,文中提出了基于改进ORB-FLANN(Oriented FAST and Rotated BRIEF-Fast Library for Approximate Nearest Neighbors)的工件图像识别方法。对ORB算法特征描述、图像特征匹配算法进行修改,解决传统图像识别算法在图像存在尺度和旋转变换情况下存在的弊端并降低误匹配率。该方法对ORB算法检测到的特征点采用SURF(Speeded Up Robust Features)算法添加方向信息并完成特征描述,得到旋转尺度不变性的特征点,结合FLANN算法并引入双向匹配策略进行特征点粗匹配,最后利用渐进采样一致算法进一步剔除误匹配点对完成精匹配。实验结果表明,与其他方法相比,改进算法在处理尺度、旋转等变换图像时,匹配正确率分别提高了2.6%~18.8%和29.5%~43.9%,运行时长均在4 s以内,提高了对工件图像的识别效率和精准性。展开更多
基金Supported by the National Natural Science Foundation of China(No.60802045)the Fundamental Research Funds for the Central Universities(No.2009JBM020)the Strategy Alliance of Chinese Academy of Sciences for Guangdong Province(No.2010B090301014)China
文摘This paper proposes an enhanced Edge Matching Rate (EMR) to gain good image regis-tration based on Generalized Acreage (GA). Traditional EMR considers only matched pixels sum without concerns of the cause of unmatched pixels and the relationship between matched pixels. The modified EMR introduces the new concept of generalized acreage to measure the overlaying parts between the target image and the model. It also defines similarity of local occlusion and of local dithering to measure interference degree. Not only edge points are considered but also non-edge points, occlusion, and dithering. Using the same preprocessing, the experiments match images based on tra-ditional EMR and the proposed EMR separately. Based on the proposed EMR the paper achieves more stable registration and higher precision.
文摘针对传统图像识别算法匹配正确率低、运行时间较长等问题,文中提出了基于改进ORB-FLANN(Oriented FAST and Rotated BRIEF-Fast Library for Approximate Nearest Neighbors)的工件图像识别方法。对ORB算法特征描述、图像特征匹配算法进行修改,解决传统图像识别算法在图像存在尺度和旋转变换情况下存在的弊端并降低误匹配率。该方法对ORB算法检测到的特征点采用SURF(Speeded Up Robust Features)算法添加方向信息并完成特征描述,得到旋转尺度不变性的特征点,结合FLANN算法并引入双向匹配策略进行特征点粗匹配,最后利用渐进采样一致算法进一步剔除误匹配点对完成精匹配。实验结果表明,与其他方法相比,改进算法在处理尺度、旋转等变换图像时,匹配正确率分别提高了2.6%~18.8%和29.5%~43.9%,运行时长均在4 s以内,提高了对工件图像的识别效率和精准性。
基金supported by the National 111 Project of China(Grant No.B17050)China Ministry of Education Project of the Humanity and Social Science Research Foundation(Grant No.19YJC790150).