计算机工程与应用 ›› 2018, Vol. 54 ›› Issue (7): 206-212.DOI: 10.3778/j.issn.1002-8331.1710-0195

• 图形图像处理 • 上一篇    下一篇

快速特征提取与感知哈希结合的图像配准算法

宋  博1,姜万里2,孙  涛1,熊正强1,芮华建1   

  1. 1.武汉大学 电子信息学院,武汉 430072
    2.陆军军官学院,合肥 230031
  • 出版日期:2018-04-01 发布日期:2018-04-16

Image registration algorithm based on fast feature extraction and perceptual hash

SONG Bo1, JIANG Wanli2, SUN Tao1, XIONG Zhengqiang1, RUI Huajian1   

  1. 1.School of Electronics and Information, Wuhan University, Wuhan 430072, China
    2.Army Officer Academy of PLA, Hefei 230031, China
  • Online:2018-04-01 Published:2018-04-16

摘要: 经典的特征点提取算法是从整个图像进行遍历来确定特征点,运算量较大,不能满足实时应用的要求。提出了一种特征点快速稀疏提取算法,该方法首先利用高斯拉普拉斯算子(Laplacian of Gaussian,LoG)提取图像梯度,设定阈值过滤获得图像的边缘稀疏矩阵,然后在稀疏矩阵的基础上利用改进的加速分割测试特征(Features from Accelerated Segment Test,FAST)检测算法,解决了传统匹配算法提取特征点耗时的问题,使图像实时匹配成为可能。为减少误匹配对,利用感知哈希算法对匹配对进行提纯,并根据仿射不变性建立两个约束条件进一步验证单应性矩阵,提高配准精度。实验结果证明,该算法提高了特征点提取的速度以及配准精度。

关键词: 稀疏矩阵, 加速分割测试特征, 特征点提取, 高斯拉普拉斯算子, 感知哈希, 仿射不变性

Abstract: There are still some problems with the classical feature detection algorithms. Traversing each pixel in image will cost much time which can’t satisfy the real time requirement. A method of fast feature detection with image sparse matrix is proposed. Firstly, the Laplacian of Gaussian(LoG) operator is used to extract the image gradient and obtain the edge sparse matrix. Then, the advanced algorithm of Features from Accelerated Segment Test(FAST) is used on the basis of sparse matrix to choose feature point, which solves the problem of time-consuming in feature detection. In addition, in order to correct mismatches that disturb image registration, perception Hash algorithm is utilized to purify the matching point pair. Finally, the homography matrix is optimized and checked by two constraints established with affine invariance. Experiments show that the algorithm improves the speed of feature extraction and the accuracy of image registration.

Key words: sparse matrix, Features from Accelerated Segment Test(FAST), feature extraction, Laplacian of Gaussian(LoG), perception Hash, affine invariance