| 摘要: |
| 目的 针对三维重建过程中尺度不变特征转换( Scale Invariant Feature Transform,SIFT) 算法对噪声敏感,导致
特征点提取和匹配的错误和运行时间长等问题,提出一种改进的 SIFT 算法,旨在提高特征点提取的准确性和减少
运行时间。 方法 改进的 SIFT 算法首先对图像的像素点进行遍历,对于每个目标像素点,将其与其 8 邻域内的像素
点进行灰度值比较。 如果相邻像素点的灰度值与目标像素点的灰度值之差小于设定的阈值,则将该相邻像素点标
记为相似点;根据相似点的数量,确定目标像素点是否为兴趣点,如果相似点的数量满足特定条件,则将目标像素
点判定为兴趣点,然后在以兴趣点为中心的区域内使用 SIFT 算法提取特征点。 结果 在不同的阈值设置和对不同
尺寸图像进行对比实验中,结果显示改进的 SIFT 算法相较于传统的 SIFT 算法,在特征点提取正确率上有约 10%
左右的提升,运行时间节约 25%左右。 结论 实验结果表明:本文提出的改进 SIFT 算法通过引入对噪声的抑制和对
兴趣点的筛选,能够有效提升特征点的提取质量,以及特征点提取和匹配中的错误率,并且显著降低运行时间。 |
| 关键词: 特征点提取 SIFT 三维重建 特征点匹配 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Research on Three-dimensional Reconstruction Technology Based on Improved SIFT Algorithm |
|
LI Ran YANG Chaoyu
|
|
School of Artificial Intelligence Anhui University of Science and Technology Anhui Huainan 232001 China
|
| Abstract: |
| Objective In response to the sensitivity of the scale invariant feature transform SIFT algorithm to noise
during the three-dimensional reconstruction process leading to errors in feature point extraction and matching as well as
long runtime an improved SIFT algorithm was proposed to enhance the accuracy of feature point extraction and reduce
runtime. Methods The improved SIFT algorithm first traversed the pixels of the image. For each target pixel it
compared the grayscale values with those of its eight neighboring pixels. If the difference in grayscale values between
adjacent pixels and the target pixel was less than a specified threshold the adjacent pixel was marked as a similar point.
Based on the number of similar points whether the target pixel was an interest point was determined. If the number of
similar points met specific conditions the target pixel was determined as an interest point and then the SIFT algorithm
was used to extract feature points within the region centered on the interest point. Results In experiments comparing
different threshold settings and images of different sizes the results indicated that the improved SIFT algorithm achieved
an approximate 10% increase in feature point extraction accuracy and saved around 25% of runtime compared with the
traditional SIFT algorithm. Conclusion The experimental results demonstrate that the proposed improved SIFT algorithm
effectively enhances the quality of feature point extraction by introducing noise suppression and interest point filtering
reducing the error rate in feature point extraction and matching and significantly reducing runtime. |
| Key words: feature point extraction SIFT three-dimensional reconstruction feature point matching |
