.png){kind=icon}
.png){kind=icon}
Return
A corner detection method based on adaptive multi-directional anisotropic diffusion
.png){kind=icon}
PRE
AI
DOI：10.1007/s11042-022-12666-w.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
As the most significant feature, corner is widely used in many application areas of computer vision, such as object tracking, image restoration, and 3D reconstruction. Various noises in the image bring about non-negligible negative impact on the accuracy of feature location for a corner detection algorithm. To avoid the influence of noise, the Gaussian filter was utilized by existing algorithms, which lead to loss of detailed information of image edges or even loss of corners. Considering the geometric structure of the corners, a new anisotropic diffusion method taking into consideration the image local multi-directional information was designed at first to achieve significant denoising effect and preserve the edge information and detailed information of the image. Subsequently, a multi-directional structure tensor product is applied to construct feasible corner measure function for detecting corners with high robustness. Finally, metrics about location accuracy, average repeatability, and image matching performance were applied to evaluated the performance of proposed corner detection method. Compare with twelve state-of-the-art methods, the experiments show that the proposed method is optimal in overall performance and the average score is 0.8504. Comparing with other methods, the proposed method has 1%-24% improvement in average performance with image affine transformation. The corner location error is 1.2216 on 'Lab', 1.2617 on 'Block' and 1.7002 on 'Pentagon', which are better than other detectors. In experiment with light and viewpoint changes, our proposed method outperforms other methods by 2.7% to 35.76% on average matching score.
Keywords:
Corner detection
Anisotropic diffusion
Multi-direction information
Image matching
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W
Organization
No organization information available