Stitching partially overlapping image tiles into a montage is a common requirement for materials microscopy. We developed ITKMontage, a new module for the open-source Insight Toolkit (ITK), capable of robustly and quickly generating extremely large, high-bit-depth montages within the memory constraints of standard workstations. The phase correlation method is at the core of our pairwise tile registration algorithm. Precise alignment of tiles acquired in typical raster patterns is enhanced with sub-pixel fitting and cropping to overlap. Fast Fourier transform (FFT)-based correlation is improved through a variety of padding methods and an added adjustable bias toward an expected translation. To arrange the tiles into the overall montage, we use global least squares minimization with outlier detection and removal. To blend tiles smoothly, each tile’s contribution is weighted by the distance from the tile’s edge. Results are demonstrated on several material science data sets and 3D images. Results compare favorably to ImageJ/Fiji plugin Image Stitching. The tool is integrated into the DREAM.3D software suite for multidimensional, multimodal microstructural data.

将部分重叠的图像拼贴缝合成蒙太奇是材料显微镜检查的普遍要求。我们开发了ITKMontage，这是开源Insight工具包（ITK）的新模块，能够在标准工作站的内存限制内快速，强大地生成超大型，高位深度的蒙太奇文件。相位相关方法是我们的​​成对图块配准算法的核心。通过子像素拟合和裁切以使其重叠，可以增强在典型光栅图案中采集的图块的精确对齐。通过多种填充方法以及针对预期转换的可调整偏差，改进了基于快速傅里叶变换（FFT）的相关性。为了将图块排列成整体蒙太奇，我们使用全局最小二乘最小化以及离群值检测和去除。为了平滑地混合图块，每个图块的贡献均按距图块边缘的距离加权。结果在几个材料科学数据集和3D图像上得到证明。结果优于ImageJ / Fiji插件图像拼接。该工具已集成到DREAM.3D软件套件中，用于多维，多峰微结构数据。