基于灰度与角点检测的图像匹配算法的研究与实现
摘 要图像配准是对取自不同时间、不同传感器或者不同视角的同一场景的两幅图像或者多幅图像进行匹配的过程，随着计算机视觉技术的发展，图像匹配技术被广泛地应用在遥感图像、医学图像、三文重构、机器人视觉等诸多领域中。图像配准的方法主要有基于像素的图像配准方法和基于特征的图像配准方法。其中基于特征的图像配准方法主要是通过计算图像每个点处的角点响应函数，选取一个区域内响应函数值最大的点作为Harris角点，或采用基于尺度不变特征变换理论的SIFT角点检测方法获取图像的特征点，最后通过奇异值分解法来进行特征点的匹配，实验证明该方法对于两幅图像间的角点有较佳的匹配效果。而对于传统的基于灰度的模板匹配算法，由于其简单实用性，得到较广泛的应用。同时由于其匹配过程需要对图像的全部灰度信息来进行配准，虽然配准精确，但是由于计算量大，需要耗费过多的CPU时间，故其在实际的应用中有较大的局限性。本文通过采用随机化的方法来获取一个较精确的阈值，减少匹配过程的无效匹配时间，经过对比实验表明，该方法在保证匹配精度的前提下，可以有效地减少匹配时间；而采用动态规划的思想通过保存所有区域灰度值的累加和来快速获取特定一块区域的灰度值的累计和的方法可以从算法上降低其时间复杂度，试验证明该方法在确保了模板匹配的优点上，可以从根本上减少了计算量，大大减少了匹配过程所花费的时间，能够满足实际生产应用的实时性要求。论文网http://www.youerw.com/
关键字：图像匹配；角点检测；Harris角点；SIFT；SVD；模板匹配
Abstract本文来自优,文~论^文·网原文请找腾讯324'9114
Image registration is the matching process of the two or multiple images in the same scene taken from different time, different sensors or different perspectives. With the development of computer vision technology, image matching technology is widely used in remote sensing images, medical images, three-dimensional reconstruction, robot vision and many other areas. The methods of image registration are mainly pixel-based image registration method and feature-based image registration method. The test, namely through calculating angular point response function of each image points, selecting the maximum value of response function in a region as a Harris angular point, or adopting the test method of the SIFT angular points based on scale invariant feature transform corner theory to obtain image feature points, and finally conducting the matching of the feature points by singular value decomposition(SVD) to the feature points, proves that the method has a much better matching results for the angular points between the two images. The traditional template matching algorithm based on gray level, because of its simpleness and practicality, is widely used. Meanwhile it needs to registrate all the gray level information of image during matching process. Although registration is accurate, the calculation is huge which requires to consume too much CPU time. So the traditional method has great limitations to its practical application. This paper tells that the randomization approach to obtain a more accurate threshold can reduce the invalid matching time. Through comparative experiments, it shows this method can effectively reduce the matching time on the premise of ensuring the accuracy of matching. And adopting the dynamic programming ideology, namely by saving the accumulated sum of gray level in all regions to quickly obtain the cumulative sum of a region-specific gray value, can reduce the time complexity in calculation. Evidence shows that the method, besides ensuring the advantages of template matching, can radically decrease the computation, greatly reduce the time spent in the matching process, and meet the real-time demand in practical application.
Key words：image matching; angular point detection; Harris angular point; SIFT; SVD; template matching,2338