Key words: dark-field imaging, sub-aperture stitching, linear optimization, error accumulation
Abstract: In the detection of large optical components, the sub-aperture stitching strategy can expand the detection range dynamically without reducing the resolution. Sub-aperture images can be matched at adjacent positions by feature matching. However, in dark-field detection for fine optical components, it is difficult to obtain correct matching results because adjacent images are always short of detectable features. In this paper, the results of feature matching are converted into the linear constraints of all step errors on the scanning path, and then the optimal solution of the step errors is obtained through least-square optimization. As a result, high-precision global stitching can be realized by correcting the step errors. In addition, the mean square error (MSE) based on the feature matching results is proposed to evaluate the stitching results. Experimental results demonstrate that this method can reduce the MSE to 3.4%–13.6% of the direct stitching and has strong robustness under different experimental conditions. Finally, we employ the matching results as feature-level information for global optimization, which helps mitigate the limitations of few matching features and can improve the accuracy and reliability of large aperture stitching tasks.