Research on double cubic interpolation combined with Gaussian fitting sub-pixel edge detection method

With the rapid advancements in intelligent manufacturing technology, the realization of fast and accurate measurements in aerospace, instrumentation, machinery and equipment, automotive electronics, and other sectors has become a research hotspot. Image positioning technology and edge detection methods, with their advantages of being fast, non-contact, and precise, have been widely employed. How to improve the accuracy of image measurement lies in solving the problem of accurate positioning of image edge detection, based on the problem of inaccurate extraction of directional information and large error in contour size of image edge detection technology proposes an improved sub-pixel edge detection algorithm with Gaussian fitting by double cubic interpolation. Based on preprocessing for pixel-level edge feature extraction, pixel-level edge points are acquired utilizing the least squares fitting algorithm. The normal section is determined by computing the normal intercept line, while the normal isometric line is identified through the Gaussian integration method. Sub-pixel fitting points are obtained via boundary fitting, and the final grey value is derived using an edge detection algorithm that combines double cubic interpolation and Gaussian fitting on the fitted points. The experimental results demonstrate that, through comparative testing of this method, it has been verified that it exhibits the highest image quality and is superior to most current algorithms. An improvement in the algorithm by 0.021%, 0.025%, and 0.031% in edge localization accuracy compared to the Zernike moment algorithm, Gaussian fitting algorithm, and bicubic interpolation algorithm has been achieved, effectively verifying that the image quality detected by this method is relatively high. In comparison with the point cloud method regarding fitting error, it was observed that the error index increased by 0.025% under strong burst light conditions, 0.026% under normal burst light conditions, and 0.022% under weak burst light conditions. The speed of processing data was doubled, further demonstrating the superiority of this method.