摘要
由于目前的SIFT(scale invariant feature transform)特征提取算法具有较高的时间复杂度,不利于大规模的数据存储和搜索,提出一种简化的SIFT局部图像特征提取算法。改进的SIFT算法针对于描述子生成部分进行简化,将原算法中特征点描述子的矩形区域改为圆形区域,并将RANSAC(random sample consensus)算法应用于SIFT特征匹配中,有效地剔除错误匹配点。采用K.Mikolajczyk的衡量方法,即查全率和错误率进行评估。实验结果显示,算法在旋转、光照、视角变化等情况下都有很好的匹配效果,并且降低了时间复杂度。
As current scale invariant feature transform （SIFT）feature extraction algorithm has high time complexity, it is not conducive to large-scale data storage and search. A simplified SIFT local image feature extraction algorithm is proposed. The simplified SIFT algorithm changes the rectangular area of the descriptor in original algorithm into sub- circular area, and random sample consensus （RANSAC） algorithm is applied to SIFT feature matching. RANSAC al- gorithm can effective eliminate the errors in matching. This paper uses the measurement method of K. Mikolajczyk, which uses the recall ratio and error rate to carry out evaluation. Experiment results show that the new algorithm is invariant for rotation, light, perspective changes, and also reduces the time complexity.
出处
《仪器仪表学报》
EI
CAS
CSCD
北大核心
2012年第10期2255-2262,共8页
Chinese Journal of Scientific Instrument
基金
高等学校博士学科点专项科研基金(20050183032)
吉林省教育厅科学基金(2009604)资助项目
