基于SIFT算法的图像匹配研究

Image Matching Based on SIFT Algorithm

图像匹配是计算机视觉、摄影测量等领域的基本问题,SIFT算法由于具有旋转和尺度不变性等优良性能被广泛使用,但也并非完美。针对SIFT算法的适用性问题展开研究,采取三组图像进行匹配,SIFT算法采用SIFTGPU的实现版本。实验结果表明,大部分情况下,SIFT算法可以得到较多可靠的同名点,但如果图像中纹理单一或者重复纹理较多以及成像时角度差异过大,容易造成SIFT算法匹配质量降低。

Image matching is a fundamental problem in computer vision,photogrammetry and other related fields.Due to its rotation invariance,scale invariance and excellent performance,SIFT algorithm is widely used,but it is not perfect.In this paper,we study the applicability of SIFT algorithm with three images as test data and SIFTGPU as runtime SIFT version.Experimental results show that in most cases,SIFT algorithm can get numerous reliable tie points,but if the image has a single or repeating texture,or the angle difference is too large when getting image,SIFT matching algorithm will result in poor quality.