Dense Correspondence Field Estimation from Multiple Images

Most optical flow algorithms assume pairs of images that are acquired with an ideal, short exposure time. We present two approaches, that use additional images of a scene to estimate highly accurate, dense correspondence fields.

In our first approach we consider video sequences that are acquired with alternating exposure times so that a short-exposure image is followed by a long-exposure image that exhibits motion-blur. With the help of the two enframing short-exposure images, we can decipher not only the motion information encoded in the long-exposure image, but also estimate occlusion timings, which are a basis for artifact-free frame interpolation.

In our second approach we consider the data modality of multi-view video sequences, as it commonly occurs, e.g., in stereoscopic video. As several images capture nearly the same data of a scene, this redundancy can be used to establish more robust and consistent correspondence fields than the consideration of two images permits.