Novel View Synthesis using Landmark-based Warping

In this paper, we propose a novel approach to generate a synthetic mid-view facial image by combining two input images captured from the left and right perspectives. Our method utilizes facial landmark detection via the MediaPipe Face Mesh model, which identifies 468 key points on the face. To overcome the model’s limitation of focusing only on the central facial region, we introduce an extension strategy that automatically adds supplementary landmarks around the head using geometric offsets. These enriched landmarks are then used to compute an average facial structure, followed by Delaunay triangulation and affine warping from both input images toward the intermediate geometry. The final output is obtained by blending the two warped images. The results show a consistent and realistic synthesis of a mid-view perspective, offering potential applications in virtual reality, facial recognition, and image-based rendering.