Computer Engineering and Applications ›› 2020, Vol. 56 ›› Issue (1): 196-202.DOI: 10.3778/j.issn.1002-8331.1809-0317

Previous Articles     Next Articles

Research on Time Series Image Mosaic Method Based on Improved SIFT

LU Peng, LU Qi, ZOU Guoliang, WANG Zhenhua, HOU Qian   

  1. College of Information, Shanghai Ocean University, Shanghai 201306, China
  • Online:2020-01-01 Published:2020-01-02



  1. 上海海洋大学 信息学院,上海 201306

Abstract: An improved SIFT algorithm is proposed for the problem of SIFT(Scale Invariant Feature Transform) algorithm which has high computational complexity and long running time. In order to solve this problem, the improved SIFT algorithm expands the range of extremum points to reduce the number of extreme points, in order to increase the speed of operation. Furthermore, the improved algorithm uses a circular window of 12 rings instead of the traditional square window and simplifies the construction of SIFT feature descriptors with generating 78-dimensional SIFT feature descriptors, which further improve the operation speed of the algorithm. For the best experimental results, the Best Bin First(BBF) method and the Random Sample Consensus(RANSAC) are used in this improved algorithm. BBF is applied in the initial registration search between feature point pairs, and RANSAC is used to perform secondary processing on the feature point registration pairs to eliminate mismatching. Finally, this paper combines the improved SIFT algorithm with the gradual image fusion algorithm to realize the mosaic and fusion of time series images. To evaluate the experimental effect, this paper uses partial block detection to evaluate after image mosaic and fusion. The experimental results show that the algorithm has fast computation speed, high robustness and good fusion effect.

Key words: Scale Invariant Feature Transform(SIFT), random sample consensus, gradual image fusion, image mosaic

摘要: 针对SIFT(Scale Invariant Feature Transform)算法计算复杂度高,运行时间长的问题,提出了一种改进的SIFT算法。通过扩大极值点取值范围,减少极值点数量,提高运算速度;采用12环的圆形窗口代替传统的方形窗口,简化了特征描述符的构造方法,生成78维SIFT特征描述符,进一步提高了算法的运算速度;将BBF(Best Bin First)运用到特征点对之间初次配准的搜索中,并用RANSAC(Random Sample Consensus)算法对特征点配准对进行二次处理,以消除错误配准。将改进的SIFT算法与渐入渐出融合算法相结合,实现对时间序列图像的拼接融合处理。针对拼接融合后的图像,采用局部分块检测的方法评价其效果。实验结果表明,该算法运算速度快,具有较高的鲁棒性,且拼接融合效果好。

关键词: 尺度不变特征变换, 随机抽样一致性, 渐入渐出融合, 图像拼接