Aswin C Sankaranarayanan

Assistant Professor, ECE
Carnegie Mellon University

mail. 5000 Forbes Ave, Porter B17, Pittsburgh, PA 15213

office. Porter B17

email. saswin -at- andrew .dot. cmu .dot. edu


Our fascination with detail is never-ending. We have cameras that capture images with billions of pixels and videos at millions of frames per second. At the heart of such technologies is the simple idea that to represent a signal with more detail, we need to sample it faster and with higher resolution. Unfortunately, this idea does not extend to many applications where sensing is inherently costly and where we cannot easily build high-resolution sampling-based sensors.

My research broadly focuses on the role of signal models in breaking traditional sensing and processing limitations. My research focuses on two main topics:

  • Co-design of optics/imaging and processing for novel sensor design

  • Use of non-linear signal models for efficient sensing and processing of high-dimensional data

compressive sensing and computational photography

BlurBurst: Removing blur due to camera shake using multiple images

Video compressive sensing using spatial multiplexing cameras
ICCP 2012

Compressive acquisition of linear dynamical systems
SIIMS 2013

SpaRCS: Compressive sensing of low rank and sparse matrices
NIPS 2011

Flutter shutter video camera for compressive sensing of high-speed videos
ICCP 2012

Compressive sensing for background subtraction
ECCV 2008

Image invariants for smooth reflective surfaces
ECCV 2012

Specular surface reconstruction from sparse reflection correspondences
CVPR 2012

non-linear signal models

Learning manifolds in the wild

A convex approach for learning near-isometric linear embeddings

Greedy feature selection for subspace clustering
JMLR 2013

Optical flow-based transport on image manifolds
ACHA 2013

multi-camera systems

Pose-invariant face recognition from multi-view videos
Trans. IP*

Joint albedo estimation and pose tracking from video
PAMI 2013

Real time head pose tracking from multiple cameras with a generic model
AMFG 2010

Distributed detection, tracking and recognition using a network of video cameras
PIEEE 2008


  • NuMax.
    A linear dimensionality reduction technique that is near-isometric (aka preserves pairwise distances) on a dataset. (code) (paper)

  • CS-MUVI.
    Video compressive sensing using the CS-MUVI algorithm.
    (project page) (code) (paper)

  • SpaRCS.
    A greedy algorithm for recovering a matrix that is a sum of low rank and sparse matrices from its compressive measurements.
    (code) (paper)

  • CS-LDS.
    Video compressive sensing under a linear dynamical system model for the data. This encompasses data lying on an unknown low-dimensional subspace.
    (project page) (code) (paper)

research group


Fall 2013. Signals and Systems (18-290)

Spring 2013. Compressive Sensing and Sparse Optimization (18-799J)