欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
现有特征点方向分配算法易受噪声干扰，在光照、仿射变换时准确性有待提高。针对以上不足，在SIFT算法基础上，提出了一种改进的特征点方向分配算法。该算法以特征点为中心，在0°～360°的范围内固定角度间隔，等距采样若干局部区域的圆形图像小块，计算各圆形图像小块质心相对圆心的偏移值。根据统计学原理以及实验验证表明，低偏移值区域易受噪声干扰且对特征点主方向的确定没有影响。据此，改进算法排除低偏移值局部区域，计算剩余局部区域像素梯度的幅度与幅角，利用方向直方图给特征点分配主方向。结果表明，相比SIFT算法，改进算法在主方向分配时运行速度更快，同时准确性更高。此外，在特征点匹配实验中，对视角变换的数据集图像，改进算法的准确率与现有算法基本持平；在噪声干扰的数据集图像中，改进算法的准确率提升了17％。
英文摘要：
Existing feature point orientation assignment algorithm is easily disturbed by noise and its accuracy needs to be improved in il- lumination variation and affine transform. For this, an improved algorithm for assigning orientations of feature points is put forward based on SIFT. It fixes the interval in the range of 0 -360 degrees with the feature points as the center, equidistant sampling of circular image blocks in several local area, calculating the offset value of the centroid for circular image blocks to center of that. According to the statis- tics theory and experimental verification,low offset value area is easily disturbed and has no effect on the determination of dominant ori- entation. Thercforc it excludes the local area of low offset value, calculates the amplitude of the pixel gradient of the residual local area, and assigns the main direction to the feature points using the directional histogram. The results of experiments show that compared with SIFT algorithm,it has a relatively higher running speed and accuracy. In addition, during the feature point matching,its accuracy is basi- cally identical with the existing algorithm in database with changeable viewpoint and improves by 17% in database with noise interfer- ence.
同期刊论文项目
同项目期刊论文
期刊信息
位置：