.png){kind=icon}
返回
A Fast Image Stitching Algorithm Based on Texture Classification and Improved SIFT
.png){kind=icon}
OA
AI
DOI：10.1109/ACCESS.2024.3443111.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
A fast and improved scale-invariant feature transform (SIFT) image stitching algorithm is proposed based on texture classification to solve the problem of huge computational complexity. In the preprocessing stage, the phase correlation algorithm is used to calculate the overlapping regions of the images, and the structural similarity (SSIM) of the overlapping regions are calculated to avoid the impact caused by inaccurate calculation of the phase correlation algorithm. Meanwhile, gradient based texture classification method is used to avoid ineffective calculations in weak texture regions. A circular eight regions descriptor structure was designed in the descriptor generation stage. And the sum of gradients in five directions within each region was calculated to obtain a feature descriptor with a dimension of only 40. The time of feature point matching was reduced due to the descriptors of lower dimensions. Further, a twice matching method was proposed based on extreme value classification to reduce the time cost of feature point matching. The experimental results show that compared to existing algorithms, this algorithm has the best performance in terms of time cost and stitching quality in two datasets. Compared to the SIFT algorithm, the time was reduced by 73.24% and 47.58%, the root mean square error (RMSE) was reduced by 94.87% and 84.36%, the number of images with failed stitching has decreased by 93.35%. The proposed algorithm significantly reduces the time cost and improves the quality of image stitching. The proposed algorithm has certain application value in the field of real-time image stitching.
Keyword:
Feature extraction
Image stitching
Classification algorithms
Costs
Accuracy
Correlation
Image edge detection
Computational complexity
Scale-invariant feature transform (SIFT)
texture classification
extreme value classification
twice matching
期刊
IF:
3.6
论文数:
9.8W
被引数:
29.4W
机构
引用论文
.png){kind=icon}
Guided Locality Preserving Feature Matching for Remote Sensing Image Registration用于遥感图像配准的引导局部保持特征匹配
PREAI
分享收藏Remote Sensing Image Registration With Modified SIFT and Enhanced Feature Matching基于改进SIFT和增强特征匹配的遥感图像配准
PREAI
分享收藏Remote Sensing Image Registration Based on Modified SIFT and Feature Slope Grouping基于改进SIFT和特征斜率分组的遥感图像配准
PREAI
分享收藏
查看更多内容