Pixel equivalent is one of the important indicators for image measurement. Traditional pixel equivalent calibration requires the use of high-precision standard components

which is cumbersome and results in inaccurate results. In view of the above problems

a pixel equivalent calibration algorithm based on improved Tukey robust fitting is proposed. The CMOS camera is used to acquire circular feature calibration board images

then these images are preprocessed to obtain 49 circular feature images. Sub-pixel edges are extracted by combining the Canny algorithm with bilinear interpolation. To address the issue that the traditional least squares method for contour fitting is susceptible to outliers

morphological operations and the Tukey algorithm are introduced to fit the circular feature contours and calculate the center coordinates of each circle. Based on the principle of noise suppression via multi-image averaging

a multi-point averaging method is adopted to compute the distances between adjacent circle centers. By comparing these with the actual distances

the pixel equivalent is obtained. Measurement experiments are conducted on high-precision rectangular patterns on the calibration film of the image measuring instrument. The results show that compared to the least squares method and Tukey algorithm

the improved algorithm has a maximum measurement error of 0.098% and an average measurement accuracy improvement of 16%.