A Graphbased Framework for Image Restoration Amin Kheradmand
A Graph-based Framework for Image Restoration Amin Kheradmand, Peyman Milanfar Department of Electrical Engineering University of California, Santa Cruz 1
Motivation q With existing hand held cameras, degradations in the form of noise/blur in the captured images are inevitable. Restoration algorithm input image enhanced output 2
Motivation q Graph-based representation is an effective way for describing the underlying structure of images. kernel similarity matrix 3
Contributions q We have developed a general regularization framework based on a new definition of the graph Laplacian for different restoration problems. 4 blurry input deblurred output
Contributions q We have proposed a data-adaptive sharpening algorithm. 5 input image output image
Examples (deblurring) blurry input deblurred output 6
Examples (deblurring) blurry input deblurred output 7
Examples (denoising) noisy input denoised output 8
Examples (denoising) noisy input denoised output 9
Examples (sharpening) 10 input image output image
References [1] A. Kheradmand, and P. Milanfar, “Non-linear structure-aware image sharpening with difference of smoothing operators”, Frontiers in ICT, Computer Image Analysis, vol. 2, no. 22, 2015. [2] A. Kheradmand, and P. Milanfar, “A general framework for regularized, similarity-based image restoration”, IEEE Transactions on Image Processing, vol. 23, No. 12, pp. 5136 -5151, Dec. 2014. [3] A. Kheradmand P. Milanfar, “Motion deblurring with graph Laplacian regularization”, Digital Photography and Mobile Imaging XI conference, IS&T/SPIE Electronic Imaging 2015, San Francisco, CA. [4] A. Kheradmand, and P. Milanfar, “A general framework for kernel similarity-based image denoising, ” IEEE Global Conference on Signal and 11 Information Processing (Global. SIP), Dec. 2013, Austin, TX.
Thanks for your time. 12
- Slides: 12