IMU-Assisted Accurate Blur Kernel Re-Estimation in Non-Uniform Camera Shake Deblurring

Area of Science:

• Computer Vision
• Image Processing

Background:

• Camera shake causes image blur, a significant challenge in computer vision.
• Current patch-wise deblurring methods use linear models for blur kernel estimation, which are insufficient for nonlinear transformations.

Purpose of the Study:

• To develop an improved method for re-estimating poorly estimated blur kernels in non-uniform image deblurring.
• To leverage inertial measurement unit (IMU) data for more accurate blur kernel modeling.

Main Methods:

• A novel IMU-assisted approach for blur kernel re-estimation.
• Establishing a nonlinear transformation relationship model between blur kernels using IMU motion data.
• Employing an optimization problem that incorporates the nonlinear model and neighboring kernels for re-estimation.

Main Results:

• The proposed method effectively re-estimates poorly estimated blur kernels.
• Experimental results show superior performance compared to existing blur kernel re-estimation techniques.

Conclusions:

• IMU data is valuable for modeling nonlinear blur kernel transformations.
• The developed IMU-assisted method offers a significant advancement in non-uniform image deblurring.