Precision workpiece edge defect and dimensional detection based on sub-pixel anomaly prediction

• Eigen-decomposition of neighborhood differential matrix enhances gradient accuracy. • Projection vector analysis achieves defect detection of linear edge anomalies. • Kalman filtering with annular state transition matrix detects circular edge defects. • Validated on workpieces with accurate sub-pixel prediction and defect detection. Precision workpiece inspection requires both the effective identification of tiny edge defects along the edge contours and dimensional measurement—a crucial aspect of machine vision-based automated industrial image inspection. A precision workpiece inspection method is proposed based on sub-pixel edge defect detection in this study, where anomalous points are identified and excluded, resulting in improvement in the accuracy of dimensional measurement. Specifically, a sub-pixel edge localization method is developed, in which gradient detection is achieved by combining a selectable neighborhood window with eigenvalue decomposition, and the resulting gradient information is then integrated with centroid to achieve precise localization. Furthermore, based on the geometric consistency of the workpiece, a predictive model for sub-pixel edge anomalies is established through quantitative analysis of projections of sub-pixel points and positional deviations distance, enabling precise detection of defects in both linear and circular features. Defect detection simulation and experiment are performed to verify the proposed method, and the results demonstrate that the method detects edge defects as small as 0.001 mm 2 . After removing anomalous edge sub-pixels, the measurement accuracy of circular and linear feature workpieces improved by 43.6 % and 41.2 %, respectively, compared to the method without removal. The overall dimensional measurement error remains within the range of ±0.0027 mm, with an average error of ±0.002 mm. These results demonstrate that the method can effectively detect edge defects in precision workpieces, offering significant potential for advancing automated machine vision-based industrial inspection.