(EFFICIENT ESTIMATOR For SIGNIFICANT BLUR REMOVAL)
Li Xu Jiaya Jia
|
|
|
|
| Real Image Input | Deblurring Result | (Kernel Size: 95x95) |
Abstract
We propose an efficient and high-quality kernel estimation method based on using the spatial prior and the iterative support detection (ISD) kernel refinement to restore pictures from significant motion blur.
Our method can estimate very large blur kernels (i.e., PSFs) and remove significant blur quickly without much hand-tuning.
Downloads
Robust Deblurring Software (Windows Trial Version)
Non-blind Deconvolution Executable (Windows Command-line)
Our Deblurring Work
Large-Kernel Robust Motion Deblurring
High-Quality Iterative Optimization
| Our Captured Real Images | ||
|
Real Image Inputs |
Our Results |
Kernels |
|
|
 85x85 |
|
|
 55x95 |
|
|
 51x51 |
|
|
 31x31 |
|
Spatial Varient Blur (All input real images below are blurred non-uniformly, presented in other manuscripts) We estimate one PSF for each image. This inevitably produces errors. The results are however reasonable, indicating that reliable PSF estimation is important. |
||
|
Input |
|
Kernels |
 |
|
|
Our Result |
 |
|
Input |
||
Our Result |
 |
|
Input |
||
Our Result |
 |
|
Input |
||
Our Result |
 |
|
Input |
||
Our Result |
 |
|
Input |
||
Our Result |
 |
|
.gif){kind=small}