面向弱纹理图像的自适应多邻域特征点描述子  

Adaptive multi-neighborhood feature point descriptor for texture-less image

作　　者：高欢 唐自新[3] 唐玲 魏世民 GAO Huan;TANG Zi-xin;TANG Ling;WEI Shi-min(School of Modern Post,Beijing University of Posts and Telecommunications,Beijing 100876,China;School of Artificial Intelligence,Beijing University of Posts and Telecommunications,Beijing 100876,China;Beijing Institute of Spacecraft System Engineering,China Academy of Space Technology,Beijing 100094,China)

机构地区：[1]北京邮电大学现代邮政学院,北京100876 [2]北京邮电大学人工智能学院,北京100876 [3]中国空间技术研究院北京空间飞行器总体设计部,北京100094

出　　处：《计算机工程与设计》2024年第11期3375-3382,共8页Computer Engineering and Design

摘　　要：针对弱纹理场景中特征提取困难、特征匹配准确率低等问题,提出一种自适应多邻域结构张量(adaptive multi-neighborhood structure tensor,AMST)特征点描述子。基于多个图像邻域及其结构张量,多层次地表达图像结构信息,解决弱纹理图像的特征提取与匹配等问题;通过特征点密度自适应邻域数量,提高计算效率,利用海森矩阵,剔除不稳定特征点,增强算法实时性以及稳定性。实验结果表明,AMST算法在弱纹理图像上的匹配准确率达到99.90%以上,同时具有良好地旋转不变性,能够适应遮挡、截断等复杂场景,具备良好的鲁棒性。Aiming at the difficulty of feature extraction and low accuracy of feature matching in texture-less scenes,a feature point descriptor(AMST)was presented.Based on several image neighborhoods and their structure tensors,the descriptor could express the structure information of the image in multiple layers to solve the problem of feature extraction and matching in the texture-less scene.The number of neighborhood was adaptive by the density of feature points to improve the computational efficiency.The Hessian matrix was used to eliminate the unstable points and enhance the stability of the algorithm.Experimental results show that the accuracy of AMST on texture-less images is more than 99.90%with good rotation invariance.At the same time,it can adapt to complex scenes such as occlusion and truncation and has good robustness.

关 键 词：弱纹理 特征提取 描述子 结构张量 多邻域 自适应 海森矩阵 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]