Camera motion introduces spatially varying blur due to the depth changes in the 3D world. This work investigates scene configurations where such blur is produced under parallax camera motion. We present a simple, yet accurate, Image Compositing Blur (ICB) model for depth-dependent spatially varying blur. The (forward) model produces realistic motion blur from a single image, depth map, and camera trajectory. Furthermore, we utilize the ICB model, combined with a coordinate-based MLP, to learn a sharp neural representation from the blurred input. Experimental results are reported for synthetic and real examples. The results verify that the ICB forward model is computationally efficient and produces realistic blur, despite the lack of occlusion information. Additionally, our method for restoring a sharp representation proves to be a competitive approach for the deblurring task.
Our ICB model accurately describes parallax motion blur. By providing the depth and camera trajectory, we can fit a set of motion kernels $k_l$ and alpha-matte terms $\mathcal{A}_l$, which are used to blend the blur from different layers.
For comparison, we show results for the proposed ICB model and the convolutional Pixel-Wise Blur (PWB) model. Images in hot colormap represent the error between the ground truth and the estimation.
We can learn a sharp neural representation using only the blurred input by coupling a coordinate-based neural network with a blur formation model (PWB or ICB).
@inproceedings{torres2023parallaxicb,
title={Depth-Aware Image Compositing Model for Parallax Camera Motion Blur},
author={Torres, German F and K{\"a}m{\"a}r{\"a}inen, Joni},
booktitle={Image Analysis: 23rd Scandinavian Conference, SCIA 2023, Sirkka, Finland, April 18--21, 2023, Proceedings, Part I},
pages={279--296},
year={2023},
organization={Springer}
}