Projection model-driven image stitching: a novel warping method using epipolar displacement field

Jian Yu, Feipeng Da

Abstract: The paper introduces a novel image stitching method driven by the projection model, utilizing epipolar displacement field (EDF) to ensure geometric consistency in panoramic images. By leveraging the principles of epipolar geometry and infinite homography, this method ensures alignment accuracy and maintains global projectivity across stitched images. The process begins by establishing a pixel warping rule in epipolar geometry through the infinite homography. Then, the epipolar displacement field, which quantifies the displacement of pixels along the epipolar lines, is constructed using thin-plate splines derived from the principle of local elastic deformation. The final panoramic image is produced by inversely warping pixels according to the epipolar displacement field. This method integrates epipolar constraints into the warping rules, ensuring high-quality alignment and preserving projection accuracy. Comparative experiments, both qualitative and quantitative, show that the method effectively reduces parallax artifacts and maintains geometric fidelity, achieving an average SSIM of 0.928 and PSNR of 29.900 across 12 scenes, outperforming existing techniques.