Abstract:

In automatic registration of high resolution imagery, the main problem was how to select templates or subimages, which were noise free and had strong features that can aid in the registration, and how to reduce search data and space. This paper provided a description of the method and flow chart of an automatic and subimage selection and presented a method of gradient energy-based selection and subimage selection with more noise and dynamic objects such as cars and trees which was easier to change in different temporal image. Multi-resolution wavelet decomposition techniques and self-adaptive image matching are used to reduce search space and data. Test results with high spatial imagery show good performance.

Key words:  high resolution imagery, multi-template matching, image automatic registration.

摘要：

提出一种基于分层多模板匹配的高空间分辨率影像的自动配准方法,并根据不同季相遥感影像的特点提出稳定性模板自动选取的方法和流程,详细描述了基于梯度能量的模板选取方法,以及如何筛选容易受到噪声、动态目标干扰的模板区域,如基于植被覆盖度的方法筛选可能受到植被季相变化的模板。同时,还运用了小波多尺度图像分解和自适应匹配的方法来减小模板匹配中数据计算量和搜索空间。通过模板匹配的精度和最小二乘法对匹配后的模板进行筛选,建立图像之间的影射关系,从而实现遥感影像的自动配准。

关键词: 高分辨率影像, 多模板匹配, 图像自动配准