Dense Correspondence Estimation for Image Interpolation

We evaluate the current state-of-the-art in dense correspondence estimation

for the use in multi-image interpolation algorithms. The evaluation is carried

out on three real-world scenes and one synthetic scene, each featuring varying challenges for dense correspondence estimation. The primary focus of our study is on the perceptual quality of the interpolation sequences created from the estimated flow fields. Perceptual plausibility is assessed by means of a psychophysical user study. Our results show that current state-of-the-art in dense correspondence estimation does not produce visually plausible interpolations.