SA-LIFT: a similar area learning invariant feature transform network framework
Nanyan Liu, Zongfa Shen · 2021
Image stitching adopts image registration and image fusion technology, which combines several images with overlapping areas into a wide-view, high-resolution panoramic image. Image Stitching is widely used in remote sensing image processing, medical diagnosis, military detection and many other fields. When the image is stitched, the feature points extracted by methods based on grayscale, frequency domain and image features are too redundant, and the feature points are not comprehensive enough. In order to solve the above problems, this paper proposes a network framework for extracting feature points of image stitching based on deep learning, which is called SA-LIFT. First, the image is divided into similar and non-similar areas by extracting model. Second, an improved LIFT model is proposed, which uses three convolutional networks to extract image feature points. The experimental results show that the method used in this paper has good effects in improving comprehensiveness and reducing redundancy when extracting the feature points of image stitching.