| CAO Hengrui*(曹恒睿),QIN Danyang* **,ZHANG Xiao* **,CHEN Yuhong* **,ZHAO Bojia*.[J].高技术通讯(英文),2025,31(2):144~153 |
|
| Research on feature matching strategy based on adaptive affine transformation |
| |
| DOI：10. 3772 / j. issn. 1006-6748. 2025. 02. 004 |
| 中文关键词: |
| 英文关键词: adaptive affine transformation, optimizing the longitude angle sampling, weak texture scene, large angle tilting |
| 基金项目: |
| Author Name | Affiliation | | CAO Hengrui*(曹恒睿) | (* Key Lab of Electronic and Communication Engineering, Heilongjiang University, Harbin 150080, P. R. China)
(** National Mobile Communications Research Laboratory, Southeast University, Nanjing 210096, P. R. China) | | QIN Danyang* ** | | | ZHANG Xiao* ** | | | CHEN Yuhong* ** | | | ZHAO Bojia* | |
|
| Hits: 34 |
| Download times: 45 |
| 中文摘要: |
| |
| 英文摘要: |
| The inherent challenges arising from variations in user-captured viewpoints and object orientation disparities in real-world scenarios pose significant difficulties in establishing robust correspondence relationships between image pairs. Methods based on geometric transformation estimation usually perform affine transformation of the global image for viewpoint correction, which not only increases the time complexity but also generates a large number of redundant features. To solve this problem, this paper proposes an adaptive affine transformation model (AATM) to achieve robust image matching by dividing special regions with pixel information and employing feature extraction algorithms with different granularities. First, the input image is divided into significant and non-significant regions by an adaptive algorithm. Second, for the salient region, the feature point extraction is accelerated by optimizing the longitude angle sampling algorithm and constructing the affine invariant nonlinear scale space, introducing the Hessian integral image and box filter. Then, for the non-significant region of the weak texture scene through the uniform step sampling algorithm, a dense feature description can be obtained in the weak texture scenes, so that more robust features are extracted for both significant and non-significant regions. The results of extensive experiments on two datasets show that the AATM algorithm outperforms similar algorithms in terms of the number of correctly matched pairs, elapsed time, and root mean square error (RMSE), indicating that the AATM can obtain more robust matches in scenes with large angle tilting and scale transformations. |
|
View Full Text
View/Add Comment Download reader |
| Close |
|
|
|
