Skip to content
Conference

Instance segmentation of shot-peening pits on textured metal surfaces using photometric stereo geometric features

Oct 2026 · Conference on Advanced Algorithms and Signal Image Processing · Vol 14370, pp. 143701S - 143701S-8 · 0 citations · 12 references
Engineering

Abstract

Shot peening is a common process for surface strengthening and residual-stress conditioning of metal parts, and the number, size, and spatial distribution of the resulting pits are key indicators of peening quality. On metal surfaces that combine strong directional machining texture with high specular reflectance, the brightness of a single grayscale image is affected at once by illumination, reflectance, texture, and geometry, so direct thresholding or edge detection easily produces texture false positives, missed pits, and merged neighboring instances. This paper proposes a traditional-vision method for instance segmentation of shot-peening pits based on photometric-stereo geometric features. Taking the per-pixel normal field reconstructed by photometric stereo as input, the method builds a slope-divergence map (div) from the slope field to enhance the two-dimensional closed geometric response at pit boundaries while suppressing directional texture; a difference-of-Gaussians (DoG) operator scale-enhances div to stabilize the pit-center response, and a dual-threshold marker-controlled watershed generates initial instances; a seed-stage shape filter removes elongated, oversized, and lowcircularity false instances while preserving small near-circular pits; finally a low-valued closed divergence wall constrains a limited number of dilation rounds so that instances approach the true boundary, and a fixed number of free-dilation rounds compensate for the wall thickness. With fixed parameters, the method reaches F1 of 0.9713, 0.9679, and 0.9551 and mean IoU of 0.8756, 0.8783, and 0.8695 on three 850×850-pixel regions of interest. Training-free, fixed-parameter, and interpretable, the pipeline provides a feasible solution for pit-level segmentation, counting, and sizing on textured metal surfaces.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.