SUMMARY In modern microscopy, the field of view is often increased by obtaining an image mosaic, where multiple sub-images are taken side-by-side and combined post-acquisition. Mosaic imaging often leads to long imaging times that can increase the probability of sample deformation during the acquisition due to, e.g. changes in the environment, damage caused by the radiation used to probe the sample or biologically induced deterioration. Here we propose a technique, based on local phase correlation, to detect the deformations and construct an artifact-free image mosaic from deformed sub-images. The implementation of the method supports distributed computing and can be used to generate teravoxel-size mosaics. We demonstrate its capabilities by assembling a 5.6 teravoxel tomographic image mosaic of microvasculature in whole mouse brain. The method is compared to existing rigid stitching implementations designed for very large datasets, and observed to create artifact-free image mosaics in comparable runtime with the same hardware resources. AVAILABILITY AND IMPLEMENTATION The stitching software and C++/Python source code are available at GitHub (https://github.com/arttumiettinen/pi2) along with an example dataset and user instructions.

中文翻译：

---

NRStitcher：万亿像素级体积图像的非刚性缝合。

发明内容在现代显微镜中，通常通过获得图像镶嵌来增加视野，在图像镶嵌中，并排拍摄多个子图像并在采集后组合。马赛克成像通常会导致较长的成像时间，这可能会增加采集过程中样品变形的可能性，例如，环境变化，用于探测样品的辐射引起的损坏或生物诱发的劣化。在这里，我们提出一种基于局部相位相关性的技术，以检测变形并从变形的子图像构建无伪像的图像镶嵌。该方法的实现支持分布式计算，并可用于生成teravoxel尺寸的镶嵌图。我们通过在整个小鼠大脑中组装微血管系统的5.6 teravoxel断层图像马赛克来证明其功能。将该方法与专为超大型数据集设计的现有刚性拼接实现方案进行了比较，并观察到在相同硬件资源下可比较的运行时中可以创建无伪像的图像镶嵌图。可用性和实现可以在GitHub（https://github.com/arttumiettinen/pi2）上获得拼接软件和C ++ / Python源代码，以及示例数据集和用户说明。