Plenoptic Image Motion Deblurring.

Area of Science:

• Computer Vision
• Image Processing
• Computational Photography

Background:

• Light field (LF) images captured by plenoptic cameras suffer from inherent aliasing and defocus blur due to scene 3D geometry.
• Existing motion deblurring algorithms for standard cameras are not directly applicable to LF images.
• High dimensionality of LF images makes traditional iterative blind deconvolution models computationally impractical.

Purpose of the Study:

• To propose a novel blind deconvolution method for removing motion blur from single light field images.
• To address the challenges posed by the unique imaging characteristics of plenoptic cameras.
• To enable accurate recovery of sharp scene textures and camera motion from blurry light field data.

Main Methods:

• Leveraging periodic patterns in plenoptic images to develop highly parallelizable image synthesis schemes.
• Implementing efficient graphics processing unit (GPU) accelerated iterative methods for blind deconvolution.
• Formulating the deblurring problem as a regularized energy minimization to recover latent sharp images and motion parameters.

Main Results:

• Successfully removed motion blur from synthetic and real light field data.
• Demonstrated the capability to handle non-uniform motion blur caused by camera shake.
• Achieved recovery of sharp high-resolution scene textures and accurate camera motion estimation.

Conclusions:

• The proposed method effectively addresses motion blur in light field images captured by moving plenoptic cameras.
• Exploiting periodic patterns enables efficient GPU implementations for complex deconvolution tasks.
• The formulation is robust and applicable to various motion blur scenarios, including camera shake.