Brief Biography

Updated July 2014

Peyman Milanfar is a Professor of Electrical Engineering. He has been on leave since 2012 at Google [x] and Google Research. He was Associate Dean for research and graduate studies from 2010 to 2012. He received the B.S. in EE/Mathematics from Berkeley, and his Ph.D. in EECS from MIT. Prior to UCSC, he was at SRI, and a Consulting Professor of CS at Stanford. In 2005 he founded MotionDSP , which has brought state-of-art video enhancement to market. His interests are in statistical signal, image and video processing, computational photography and machine vision. He is a Fellow of the IEEE.

Recent News

Updated July 2014

  • May 2014: I gave a plenary talk at the Technion's TCE Conference . Here is the video of my talk along with the slides .
  • March 2014: New Software package for our paper Deconvolving PSFs for A Better Motion Deblurring using Multiple Images European Conference on Computer Vision (ECCV) 2012
  • December 2013: I gave a plenary talk at the 2013 Picture Coding Symposium . The slides for my talk can be found here.
  • October 2013: New patent no. 8,559,671 issued: Training-free generic object detection in 2-D and 3-D using locally adaptive regression kernels
  • October 2013: Do you use patch-based methods like BM3D and NLM? Time to upgrade! In our recent paper, we propose a global, practical method which in effect uses all the pixels in the input image to denoise every single pixel. The global approach goes beyond the dominant paradigm of non-local patch-based processing, which we have shown to be inherently limited. Pixels are denoised individually, not as patches. The global filter can be implemented efficiently by sampling a fairly small percentage of the pixels in the image, and can effectively globalize any existing denoising filters, improving upon the best patch-based methods. And unlike these, the global performance always improves with increasing image size.
  • July 2013: Here is a nice way to improve the performance of almost any denoising method: Symmetrize the matrix that defines the filter. This paper describes the why and the how of this process, and shows that it is stable, and widely applicable.

Paper Highlights

Updated July 2014

H. Talebi and P. Milanfar, "Global Image Denoising", IEEE Transactions on Image Processing, vol. 23, no. 2, pp. 755-768, Feb. 2014
The global approach goes beyond the dominant paradigm of non-local patch-based processing, which we have shown to be inherently limited. Pixels are denoised individually, not as patches. And every pixel helps denoise every other pixel. And the performance always gets better with larger images. Teaser Image
A. Kheradmand and P. Milanfar, "A General Framework for Regularized, Similarity-based Image Restoration", IEEE Transactions on Image Processing, submitted
We've developed an iterative graph-based framework for image restoration based on a new definition of the normalized graph Laplacian. We propose a cost function which consists of a new data fidelity term and a regularization term derived from the specific definition of the normalized graph Laplacian. The specific form of the cost function allows us to render the spectral analysis for the algorithm. The approach is general in the sense that we have shown its effectiveness for different restoration problems including deblurring, denoising, and sharpening. Teaser Image
P. Milanfar, " A Tour of Modern Image Filtering ", IEEE Signal Processing Magazine vol. 30, no. 1, January 2013, pp. 106 -128
This is my "treatise" on some key intersecting ideas in image processing, machine vision, learning, and graphics that have defined the modern era of adaptive processing using non-parametric methods.