Integrating polarization imaging and CIELCH color metrics for robust defect inspection on complex surfaces

To address challenges in detecting defects on complex curved surfaces such as reflections, minute color variations, and diverse shapes, this paper proposes a high-precision visual inspection method combining polarization imaging with CMC color difference quantification in the CIELCH color space. By fusing multiband frequency domains to reconstruct backgrounds and incorporating the Kirsch operator, it achieves precise extraction of concave-convex defects and background restoration. For thermos cup inspection, this method achieves a 96.02% defect recognition rate for exposed base-color defects, with a false-positive rate of only 4.12%. Compared with YOLOv8, recognition accuracy improves by approximately 43%, while the false-positive rate for small-area defects drops from 55% to 4%. Furthermore, the texture extraction combining multiband fusion and the Kirsch operator achieved a Dice coefficient exceeding 0.94, surpassing traditional methods by over 8%. This effectively enables highly reliable inspection of products with complex curved surfaces.