Deblatting (deblurring and matting) of fast moving objects

Restoration of motion-blurred objects. From the given input image containing the blurred object in front of a static background and a separate background image the task is to estimate the shape and restore the sharp appearance of the object as well as its motion blur (trajectory). At its current state the approach is limited to 2D objects and 2D motions, although some preliminary experiments with 3D have been made (see e.g. Rozumnyi CVPR20).

Publications:

  1. Jan Kotera, Jiří Matas, Filip Šroubek: Restoration of fast moving objects, IEEE Transactions on Image Processing (under review), [2020]
  2. Jan Kotera: Image Deblurring in Demanding Conditions, doctoral thesis, [2020] Download
  3. Kotera Jan, Šroubek FilipMotion Estimation and Deblurring of Fast Moving Objects, Proceedings of the 25th IEEE International Conference on Image Processing 2018, p. 2860-2864, ICIP 2018, [2018]

 

Example input images:

 

Image acquisition model:

where

 

Selected results:

  • Performance on the dataset in [1]

  • Results on the dataset in [1]

  • Deblurring of cars

  • Real-world free-flying objects

  • Temporal super-resolution of a video sequence with a fast moving object