Motion Blur Datasets and Matlab Codes
Agrawal, Yi Xu, Ramesh Raskar and Jack Tumblin
We provide datasets for evaluating different motion blur algorithms and capture procedures. These datasets include (a) high speed videos, (b) coded exposure images and (c) varying exposure images. All datasets are captured using a static camera.
Codes used in Coded Exposure:
A. Optimal codes for coded exposure for code length up to 100. Each of these codes are 50% on/off. For smaller code lengths (<=24), all possible codes can be tested to find the optimal code. For larger code lengths, we randomly search 1 million codes. The criteria of optimality is to maximize the minimum of the DFT of zero padded codes.
B. Matlab/C build to search for optimal code of length n for coded exposure.
Coded Exposure Deblurring (Matlab Code)
The matlab code shows the correct way of deblurring coded exposure images. Note that when the blur size is larger than the code length, deblurring does not result in deconvolution artifacts. Only the minimal resolvable blur size is increased. One should not see any ghosting or deconvolution artifacts.
We hope these datasets would be useful to students and researchers working in this field. Using high speed videos, several cameras can be simulated easily. For example, a traditional camera with a finite exposure can be simulated by simply averaging frames. A coded exposure camera can be simulated by adding frames according to the code. Camera motion can be simulated to a large extend by shifting the images according to camera motion before averaging. The datasets include high speed videos of a moving ISO resolution chart, which will be useful to evaluate the quality of deblurring algorithms/capture procedures. Noise analysis can be done using the homogeneous parts of the resolution chart. (See our CVPR 2009 paper for more details).
Dataset A: High Speed Videos
(Click on images to download .zip file of frames and associated matlab code)
Dataset B: Coded Exposure Images using Canon Camera and Ferro-Electric Shutter
(Download complete matlab code and all four input files)
Dataset C: Coded Exposure Images using Pointgrey Dragonfly2 Camera (Trigger mode 5)
Dataset D: Varying Exposure Images using Canon Rebel XT (AEB mode)
(Download images directly)
Dataset E: Varying Exposure Images using PointGrey Flea2 Camera
1. Matlab Code
for Deblurring Coded Exposure Images in
SIGGRAPH 2006 paper
Joint work with Yi Xu, Ramesh
Raskar and Jack
Related Papers in Motion/Focus Deblurring from our
SIGGRAPH 2006 Coded exposure for motion
Permission to use, copy and modify softwares linked from this page and its documentation without fee for educational, research and non-profit purposes, is hereby granted, provided that the above copyright notice and the following three paragraphs appear in all copies.
To request permission to incorporate this software into
products contact: Vice President of Marketing and Business
Mitsubishi Electric Research Laboratories (MERL), 201
Cambridge, MA 02139 or
MERL SPECIFICALLY DISCLAIMS ANY WARRANTIES, INCLUDING,
LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND
FITNESS FOR A
PARTICULAR PURPOSE. THE SOFTWARE PROVIDED HEREUNDER IS ON
AN "AS IS"
BASIS, AND MERL HAS NO OBLIGATIONS TO PROVIDE MAINTENANCE,
UPDATES, ENHANCEMENTS OR MODIFICATIONS.