Abstract:Target detection in 1-D images (grayscale curves) is a fundamental problem in machine vision measurement. Under the interference of noise, there are many difficulties in robustly estimating the position of the curve center lacking shape prior information. In this paper, a symmetry center fitting algorithm is proposed. The matching error between the curve and its mirror image is used as the symmetry evaluation function, and the least square method is used to calculate the best matching point as the symmetry center. The algorithm needs iterative calculation, and may converge to wrong position. By analyzing the possible local convergence caused by the initial value selection of iterative, it is proved that the local extreme point may only appear within the half pixel adjacent to the true value, a convergence point verification strategy is proposed, which solves the problem of erroneous convergence. The robustness of the algorithm under various disturbances is confirmed through simulation and real image verification.