摘要
针对SIFT描述符维度过高,不具备仿射不变性问题,研究人员将简化描述符与ASIFT模拟机制相结合提出一种完全仿射不变的匹配算法。首先对每一幅图像进行仿射变换模拟由于摄像机位置变化产生的仿射形变,为了模拟出尽可能多的形变,对摄像机经纬纬度采样时保证绝对倾斜按照等比数列的规律增大,经度按照等差数列的规律增大,而且公差随着绝对倾斜的增大而减小。然后检测每一幅模拟图像的SIFT关键点生成描述符,为了降低描述符的维度,提出将圆形邻域划分为4个扇形区域,在扇形区域内统计8个方向的梯度直方图形成32维描述符的方法。通过与ASIFT进行对比实验,结果表明该方法大大提高匹配速度,而且对于图像的仿射形变具有较好的鲁棒性。
Combining simplified descriptor and simulation mechanism of ASIFT, the relevant researchers proposed a fully affine invariant matching al- gorithm to address the problem that SIFT descriptor dimension is too high and does not have affine invariance. Firstly, each image was transformed by simulating affine distortions caused by changes of camera positions. When longitudes and latitudes were sampled, to effec- tively simulate as many distortions as possible, it should be guaranteed that the absolute tilts increased as a geometric series and the longi- tudes increased as a arithmetic series, in addition, the associated tolerance should be inversely proportional to absolute tilts. Secondly SIFT key points were detected on all simulation images to generate descriptor information. In order to reduce descriptor dimension, we parti- tioned a circular region into four sector bins and by computing the gradient orientation histogram in eight directions for each bin, a descrip- tor with 32 dimensions was constituted. By comprehensive comparison with ASIFT, the results indicate that the proposed method can in- crease the average matching speed significantly, in addition, it is robust to affine distortions.
出处
《现代计算机》
2017年第24期20-24,29,共6页
Modern Computer
关键词
图像匹配
仿射不变性
简化描述符
Image Matching
Affine Invariance
Simplified Descriptor
