CAJ | 학술논문

特征点匹配在图像检索、模式识别等技术中起着重要的作用.已有的匹配算法如SIFT(DoG),Harris以及SUSAN算法,虽然可以提取高质量的特征点,但是这些算法本身计算量比较大,难以将其运用于实时性要求比较高的应用中.提出一种改进的快速特征点匹配算法,采用Guoshen Yu和Jean-Michel Morel提出的全仿射方法,对局部特征点进行仿射变换并模拟摄像机成像原理,根据摄像机成像的仿射关系提取特征点并使用随机蕨类算法训练分类器,使用RANSAC去除坏点,实现对特征点的快速准确匹配.实验结果表明该方法提高了图像的匹配点数,同时降低了匹配时间.
특정점필배재도상검색、모식식별등기술중기착중요적작용.이유적필배산법여SIFT(DoG),Harris이급SUSAN산법,수연가이제취고질량적특정점,단시저사산법본신계산량비교대,난이장기운용우실시성요구비교고적응용중.제출일충개진적쾌속특정점필배산법,채용Guoshen Yu화Jean-Michel Morel제출적전방사방법,대국부특정점진행방사변환병모의섭상궤성상원리,근거섭상궤성상적방사관계제취특정점병사용수궤궐류산법훈련분류기,사용RANSAC거제배점,실현대특정점적쾌속준학필배.실험결과표명해방법제고료도상적필배점수,동시강저료필배시간.
Feature points matching plays an important role in image retrieve, pattern recognition and so on. Feature detectors such as SIFT(DoG), Harris and SUSAN algorithm are good methods which yield high quality features, however they are too compu-tationally intensive for using in real-time applications of any complexity. This paper puts forward an improved fast feature point matching algorithm. And uses the full affine method to extract the local feature points, which it was advocated by Guoshen and Jean-Michel, then uses the Random Fern to match feature points, uses RANSANC to remove dead points. The experimental results show that this method improves the image matching points, and reduces the matching time.