用于遥感图像目标快速匹配识别的改进混合溢出树算法

提出一种基于标记的混合溢出树(SHSPT)特征匹配算法,用于遥感图像的目标匹配识别。针对特征数据建立和预处理,提出了基于中心点的数据分割方法,通过定义数据密集区域的中心,舍去边缘稀疏数据,提取出分割后的数据。进行特征匹配时,使用二进制数组表示数据空间,标记分割后的特征向量数据,通过比特操作计算特征向量间的距离,缩短计算时间。最后对特征匹配方法进行改进,采用待匹配特征距离的均值代替尺度不变特征变换(SIFT)匹配算法的次临近特征距离,从而得到更多的匹配点。实验证明,基于标记的混合溢出树特征匹配算法占用内存空间比传统的混合溢出树算法减少约68%,匹配准确度与原算法接近,匹配时间平均缩短了约32.8%,解决了航天遥感图像数据量大,特征维数较高,匹配识别时间长,占用计算机内存大等问题。

Improved hybrid spill-tree algorithm for fast target matching recognition of satellite images

An improved Hybrid Spill-tree algorithm based on the signed method defined as Signed Hybrid Spill-tree (SHSPT) was proposed for target matching of remote sensing images. For establishing data and preprocessing, a data separate method based on a center point was proposed, the separated data were extracted by defining the center of dense data, and the edge data were abandoned. In the feature matching, binary array were used to express the data space and to mark the feature vector. Then, the bit operation was used to compute the distance between the feature vectors and to shorten the computing time. Finally, the feature matching algorithm was improved. The average value of the feature distance was used to replace the secondary characteristic distance from the Scale Invariant Feature Transform(SIFT)matching algorithm to obtain more matching points. The test results show that the computer memory by proposed algorithm is reduced 68% than that of traditional hybrid spill-tree method, and matching accuracy is closed to that of the traditional one. In addition, the method reduces 32.8% matching time. It solves the problems of remote sensing images in larger data amounts, higher dimensions, longer matching time and larger computer memory.