Abstract: Aiming at the problems that 128-dimensional description of the feature point reduces the efficiency of the Scale Invariant Feature Transform(SIFT) algorithm, this paper presents an improved SIFT algorithm, which uses ring and sequence of each feature vector to ensure rotation invariance, while reducing the description of operator dimension and using traversal search to find a sample of the nearest neighbor feature points and the next nearest neighbor feature points. Experimental results show that when there are different levels of image geometric distortion, radiation distortion and noise, the improved algorithm is more stable and faster.

Key words: Scale Invariant Feature Transform(SIFT) algorithm, image matching, scale invariant, feature descriptor

摘要： 针对普通SIFT算法效率因128维的特征点描述算子而降低的问题，提出一种改进的SIFT算法，利用圆环的特性同时对每一个特征向量进行序列化，以保证物体旋转不变性，在降低描述算子维数的基础上，利用遍历搜索查找样本特征点的最近邻和次近邻特征点。实验结果表明，当图像存在不同程度的几何变形、辐射畸变和噪声影响时，改进算法更稳定、更快速。

关键词: SIFT算法, 图像匹配, 尺度不变, 特征描述符

CLC Number: 

• TP311.52