Structural region-induced inpainting for highlight removal in microscopic damage measurement of metal pairs

Microscopic vision-based damage inspection is a prevalent technique in overhauling machine components. However, the image quality is often dramatically limited by the high-reflection of the metal surface during imaging, imposing obstacles to the intelligent identification of the damage region. To address this issue, a highlight removal method is proposed, centered on detecting the highlight region and the matching region, as well as constructing a removal strategy. Initially, in terms of local weak highlight noise, a two-stage detection method is developed to identify the strong highlight region to be recovered. Afterwards, utilizing a constructed structure-based clustering-growing model, the matching region around the strong highlight is identified and applied to infer the topography of the highlight region according to the local region self-similarity theory. Finally, a comprehensive priority function is established to guide an effective recovery of the highlight region in an appropriate order. For verification, damage detection was conducted using the highlight removal images and original images. Experimental results indicate that the restored highlight region exhibits a similarity ranging from 69.14 % to 98.81 % with the actual damage morphology and the damage detection accuracy of the worn surfaces is significantly improved using the proposed highlight removal method. The accuracy of damage detection is increased by 6.9 %-36.6 % on test images with pit/scratch damages, which provides a solid basis for the microscopic vision-based condition monitoring of frictional pairs.