We present a robust feature matching approach that considers features from more than two images during matching. Traditionally, corners or feature points are matched between pairs of images. Starting from one image, corresponding features are searched in the other image. Yet, often this two-image matching is only a subproblem and actually robust matches over multiple views and/ or images acquired at several instants in time are required. In our feature matching approach we consider the multi-view video data modality and find matches that are consistent in three images. Requiring neither calibrated nor synchronized cameras, we are able to reduce the percentage of wrongly matched features considerably. We evaluate the approach for different feature detectors and their natural descriptors and show an application of our improved matching approach for optical flow calculation on unsynchronized stereo sequences.

Video editing plays an important role in today’s cinematic post-production: editing operations are typically applied on a keyframe basis and propagated automatically to the rest of the sequence. Thereby, small inconsistencies in correspondences used for the propagation of editing operations accumulate and need to be corrected manually. The amount of required manual interaction increases further when editing stereoscopic video sequences. In this paper, we propose an algorithm for looped correspondence estimation yielding consistent correspondences that can be used for reliable propagation of editing operations along a stereoscopic video sequence, avoiding drift and error accumulation common to standard approaches. Taking an additional spatially adjacent image into account, we extend standard two-image optical flow algorithms and exploit data redundancy within stereoscopic video sequences. Our proposed algorithm not only yields robust and consistent correspondences for stereo footage, but also improves the accuracy of standard two-image motion estimation as demonstrated for several test scenes with known ground truth.

Traditional optic flow algorithms rely on consecutive short-exposure images. In contrast, long-exposed images contain integrated motion information directly in form of motion blur. In this project, we use the additional information provided by a long exposure image to improve robustness and accuracy of motion field estimation. Furthermore, the long exposure image can be used to determine the moment of occlusion for the pixels in any of the short exposure images that are occluded or disoccluded.
This work has been funded by the German Science Foundation, DFG MA2555/4-1

We present a method for image interpolation which is able to create high-quality, perceptually convincing transitions between recorded images. By implementing concepts derived from human vision, the problem of a physically correct image interpolation is relaxed to an image interpolation that is perceived as physically correct by human observers. We find that it suffices to focus on exact edge correspondences, homogeneous regions and coherent motion to compute such solutions. In our user study we confirm the visual quality of the proposed image interpolation approach. We show how each aspect of our approach increases the perceived quality of the interpolation results, compare the results obtained by other methods and investigate the achieved quality for different types of scenes.

The Virtual Video Camera research project is aimed to provide algorithms for rendering free-viewpoint video from asynchronous camcorder captures. We want to record our multi-video data without the need of specialized hardware or intrusive setup procedures (e.g., waving calibration patterns).