提示
双目图像特征匹配神经算子
首发时间：2025-03-14
摘要：现有立体匹配模型因在固定尺寸图像和固定视差范围下工作,存在迁移性和泛化性不足的问题。为了解决这一问题,本文提出一种新的双目图像特征匹配的神经算子方法,学习图像特征空间到视差函数空间的映射。结合基于galerkin注意力机制的gru迭代方法和构造变视差步长的cost volume,实现对不同图像分辨率和视差范围的动态调整。实验结果显示,所提方法在多个公开数据集上表现优异,具有较强的适应性和鲁棒性,为双目图像特征匹配提供了更具泛化能力的0008全讯注册的解决方案。
关键词：
for information in english, please click here
neural operators for binocular image feature matching
abstract：existing stereo matching models have limited transferability and generalization as they work on fixed - size images and within fixed disparity ranges. to solve this, this paper presents a new neural operator approach for binocular image feature matching. it learns the mapping from the image feature space to the disparity function space. by integrating the gru iterative method with galerkin attention mechanism and constructing a cost volume with variable disparity steps, it can dynamically adapt to different image resolutions and disparity ranges. experimental results show the proposed method performs well on multiple public datasets, demonstrating strong adaptability and robustness. it offers a more general - purpose solution for binocular image feature matching.
keywords：
基金：
论文图表：
引用
导出参考文献
no.****
动态公开评议
共计0人参与
勘误表
双目图像特征匹配神经算子
评论
全部评论