Image Morphing for Space-Time Interpolation

Rendering convincing transitions between individual pictures is the main challenge in image-based rendering and keyframe animation as well as the prerequisite for many stunning visual effects. We present a perception-based method for automatic image interpolation, achieving psycho-visually plausible transitions between real-world images in real-time. Based on recent discoveries in perception research, we propose an optical flow-based warping refinement method and an adaptive non-linear image blending scheme to guarantee perceptional plausibility of the interpolated in-between images. Conventional, uncalibrated photographs suffice to convincingly interpolate across space, time, and between different objects, without the need to recover 3D scene geometry, actual motion, or camera calibration. Using off-the-shelf digital cameras, we demonstrate how to continuously navigate the viewpoint between camera positions and shutter release times, how to animate still pictures, create smooth camera motion paths, and how to convincingly morph between depictions of different objects.