This paper presents a new image-based rendering method, based on photographs acquired with handheld cameras whose characteristics are unknown. Our system relies on recent calibration advances that use robust shape descriptors such as SIFT or SURF, associated with matching methods, producing a cloud of 3D points that belongs to the surface of acquired objects. The proposed method provides an interactive navigation system, with a high refresh rate, without any polygonal geometric reconstruction of objects. For a new viewpoint, our method performs the following operations: (i) point cloud projection onto the camera, (ii) depth interpolation in image space, (iii) reverse projection of input photographs to estimate the value of each pixel in the new view, (iv) occlusions management system for avoiding re-projection errors depending on the view. With this system, high dynamic range images can also be produced from freehand shooting.
展开▼
