Aero-engine blades, as one of the core high-precision components of the engine, require rapid and accurate surface measurement to ensure engine performance. Accurate extraction of laser stripe centerlines is a critical step in three-dimensional measurement, directly affecting measurement accuracy. To address the limitations of existing center extraction methods under uneven stripe width, high curvature, and noisy conditions, this study proposes a centerline extraction algorithm combining normal-guided extremum method with an improved spatial grayscale centroid propagation approach to optimize extraction precision. The method first estimates the centerline position using the Steger method; potential key feature regions are then identified via the extremum method, and weighted spatial grayscale centroid calculation along the normal direction is applied to obtain the initial stripe center. An improved eight-neighborhood angular selection is subsequently used to screen effective points, and outliers are removed using the 3σ criterion, ultimately yielding precise center coordinates. Comparative experiments demonstrate that the algorithm achieves a root mean square error (RMSE) of 0.058 pixel on standard test images, effectively preserving stripe details; 1 frame processing time is approximately 0.755 ms, achieving a 6-fold speed improvement over the Steger method and a 3-fold improvement over the conventional grayscale method, effectively addressing the curvature-induced centerline tilt problem. Furthermore, tests on blurred samples and images with added Gaussian noise show a 3.7% increase in fitting accuracy and a 6.9% decrease in RMSE compared with traditional methods, indicating superior robustness and practical applicability. The proposed algorithm provides effective technical support for high-precision measurement applications in aerospace manufacturing and optical precision instruments.