An automatic image registration evaluation model on dense feature points by pinhole camera simulation

Evaluating image registration methods in images with dense feature points is challenging, because it's hard to discriminate real inliers in thousands of resulting correspondences processed by image registration methods. Therefore, this paper presents a dense feature points simulation which could provide ground truth of correspondences for evaluating image registration methods automatically. Moreover, the dense feature points are created by simulating pinhole camera model, parallax of reference image and sensed image, radial distortion of camera lens and random outliers. The performance of five state-of-art image registration methods is evaluated and discussed on the dense correspondences simulated by proposed model. The evaluation results show that the proposed model indeed offers a practical way for evaluating image registration methods on dense feature points.