📄 中文摘要
该研究提出了一种通过观察图案在旋转和位移时的运动来识别纹理的新方法。这种方法不仅依赖于像素的比较，还通过深度网络在旋转和滑动后比较图像的各个部分，保留了方向和位置的关联信息，使得整体模式更易于识别。系统能够从单张照片中学习到稳定的信号，即使在图像旋转或缩放的情况下也能区分相似的表面。此外，特定的滤波器能够同时跟踪移动和旋转，使得该方法在不同尺寸下表现出色，并在困难的测试中比传统方法提供了更高的准确性。该方法简单易用，同时保留了细节信息。
刚性运动散射用于纹理分类
出处: Rigid-Motion Scattering for Texture Classification
发布: 2026年2月14日
📄 English Summary
Rigid-Motion Scattering for Texture Classification
A new method for texture recognition is introduced, focusing on how patterns move during rotation and shifting, rather than just pixel comparisons. A deep network analyzes parts of images after they have been rotated and shifted, preserving the information about direction and position while making overall patterns easier to identify. The system learns stable signals from a single image, allowing it to differentiate similar surfaces even when they are rotated or scaled. Additionally, specialized filters track both movements and rotations, resulting in robustness across various sizes. This approach has demonstrated higher accuracy than traditional methods in challenging tests, and it is user-friendly while retaining fine details.