This paper introduces a novel bidirectional bias correction scheme for gradient-based shift estimation (GBSE), which addresses the systematic bias caused by noise and aliasing in image registration. The method improves the accuracy of shift estimation by computing a bias matrix from multiple subpixel shifts and correcting for estimation biases that are linearly proportional to the true shift. Experimental results demonstrate that the proposed approach effectively reduces bias and enhances precision, making it suitable for various computer vision applications.