Image stitching method for defect detection on optical surface based on adaptive registration and fusion

The large-aperture optical elements are widely applied in various fields, but surface defects of the optical elements will reduce the system performance. Defect detection is one of the main interests of optical measurement research. To achieve defect detection of large-aperture elements, the defect images need to be stitched. This paper proposes a defect image stitching method for large-aperture optical elements based on the adaptive dimensionality reduction registration (ADRR) algorithm and the gradual weighted fusion (GWF) algorithm. The ADRR algorithm adaptively reduces the dimensionality of feature point descriptors extracted by the popular scale-invariant feature transform (SIFT) algorithm, and the low-dimensional descriptors are used to match feature points. The GWF algorithm constructs gradual weight matrices based on the column coordinates of overlapping areas, achieving seamless stitching of defect images. The experimental results show that the ADRR algorithm has a higher effective matching rate and shorter matching time than the SIFT algorithm. The GWF algorithm effectively reduces the interference caused by uneven background brightness. Therefore, the stitching method has good robustness and practicability.