SIFT与Harris提取图像特征点的分析研究-《计算机技术与发展》

文章信息/Info

Title:: Analysis and Research on Image Feature Points Extraction by SIFT and Harris

作者:: 赵悟¹; 段永璇¹; 段会川²; 肖宪翠²; 张睿¹; 岳媛¹; 孙小飞¹; 范军¹; 1．山东省医药卫生科技信息研究所，山东济南 250062; 2．山东师范大学信息科学与工程学院，山东济南 250014

Author(s):: ZHAO Wu¹; DUAN Yong－xuan¹; DUAN Hui－chuan²; XIAO Xian－cui²; ZHANG Rui¹; YUE Yuan¹; SUN Xiao－fei¹; FAN Jun¹; 1．Shandong Institute of Medicine and Health Information，Jinan 250062，China; 2．School of Information Science and Engineering，Shandong Normal University，Jinan 250014，China

Keywords:: SIFT algorithm; Harris corner detection algorithm; feature points availability; computational efficiency; feature point invari- ance

摘要:: SIFT(尺度不变特征变换)算法与Harris角点检测算法作为两种经典的图像特征点提取算法,在不同的图像处理中,两者体现出的图像特征点提取性能也不同。因此,如何选取合适的评价指标使两种算法在不同类型图像下提取特征点更高效,将对后续的研究与图像分析工作有重要意义。文中利用常用的折线特征主导的图像与光滑曲线特征主导的图像进行实验,并提出了一种指标评价法,即从特征点有效性、计算时效性、特征点相似不变性三方面,分别对SIFT和Harris算法在提取特征点上的有效性进行定量分析。实验结果表明。在保证特征点有效性、计算时效性以及特征点相似不变性一致的情况下,在折线特征主导的图像处理中,Harris角点检测算法与SIFT算法相比,图像特征点的提取更高效;而在光滑曲线特征主导的图像处理中,由于无法检测到足够数量的特征点,Harris角点检测算法与SIFT算法相比,图像特征点的提取性能相对较低。

Abstract:: Scale invariant feature transform (SIFT) and Harris corner detection algorithm are two classic image feature point extraction algorithms． In different image processing，they have their own traits in image feature point extraction． Therefore，how to select appropriate evaluation indexes to make the two algorithms more efficient in extracting feature points under different types of images，will be of great significance to the subsequent research and image analysis． In this paper，we take the commonly used polyline features dominant image and smooth curve features dominant image for experiment，and put forward a kind of index evaluation method． From the feature points availability，computational efficiency and feature points invariance，the effectiveness of SIFT and Harris algorithms in extracting feature points is analyzed quantitatively． The experiment shows that in the same way as the above three significant aspects，for polyline feature dominant images，Harris corner detection algorithm performs better than SIFT． But for smooth curved line feature dominant images，Harris corner detection algorithm is drastically degraded since too few feature points are detected．