CAJ | 학술논문

SIFT特征匹配算法的匹配能力强，但特征点中孤立点和噪声点等会导致部分特征点误匹配；不同图像间特征点的有关描述相近，也会造成两幅不同结构的图像，在提取出各自的 SIFT 特征点后相互匹配。为此，提出一种改进 SIFT 的图像特征匹配算法。该算法是在 SIFT 特征匹配的基础上，利用无监督学习方法对匹配异常点进行剔除，实现特征点的二次精确匹配。
SIFT특정필배산법적필배능력강，단특정점중고립점화조성점등회도치부분특정점오필배；불동도상간특정점적유관묘술상근，야회조성량폭불동결구적도상，재제취출각자적 SIFT 특정점후상호필배。위차，제출일충개진 SIFT 적도상특정필배산법。해산법시재 SIFT 특정필배적기출상，이용무감독학습방법대필배이상점진행척제，실현특정점적이차정학필배。
Due to the invariance of scale,rota-tion,illumination,SIFT (scale invariant feature transform)descriptor is commonly used in image matching.However,on the one hand,in practical applications the isolated point and the noise point may cause mismatching points.On the other hand, SIFT feature points record the relationship of dif-ferent scale between the feature point and around it,so easily caused the described similar between the different image feature point and it can be matched each other after extracting their feature points.In order to solve the problem,this paper proposed a feature point matching method based on SIFT algorithm,use unsupervised learning meth-ods to classify the matching points and eliminate the abnormal points,achieving the goal of the sec-ond accurate feature matching.