Raghuraman Gopalan

Search this site

Ph.D., July 2011


Research Interests: Computer Vision and Machine Learning





I'm a senior member of technical staff at the AT&T Labs-Research. I work at the Video and Multimedia Department, with a focus visual recognition problems. Prior to this, I did my Ph.D. at the University of Maryland's Center for Automation Research, under the supervision of Prof. Rama Chellappa. My dissertation addressed problems related to object recognition, from both statistical and geometric perspectives.


Contact Information:
AT&T Labs-Research
200 S. Laurel Avenue, Bldg. A
A5-4E05
Middletown NJ 07748 USA

Email: raghuram {AT} research.att.com


News:
  • Two journal articles accepted: Blur-invariant face recognition at Trans. PAMI [pdf], and Lane detection and tracking at Trans. Intelligent Transportation Systems [pdf].
  • Partial code for ICCV 2011 paper on domain adaptation available [download]
  • Slides on face detection presented at the ICCV tutorial on Looking at People available [download]

Links:  [Publications]  [CV]

Research:

My main focus has been on the following two specific aspects of object recognition,
  • Understanding the physics/ geometry of variations in objects, and developing representations invariant to them, 
  • Developing efficient learning formulations to harness the contextual information shared by objects with other 'points' in the scene.
I'm highly motivated by perceptual approaches to these problems, since the primary focus of computer vision is to understand the human visual capabilities. Hence, most of my work involves identifying strong prior information, and modeling them using tools from statistics and geometry. An outline of my projects is presented below.


Articulation-invariant Shape Representation
Representing a 2D non-planar shape invariant to 3D articulations, under no self-occlusions.
                • Approximate convex decomposition of an N-dimensional articulating shape using a robust area-based measure of convexity
                • Part-wise affine normalization to obtain an articulation invariant distance, under a weak-perspective camera model
R. Gopalan, P. Turaga, and R. Chellappa, "Articulation-invariant representation of non-planar shapes", European Conference on Computer Vision (ECCV) 2010. [paper][suppl. material] [poster] [spotlight] [dataset]





Space of Blurred Images: A Robust Blur-invariant Descriptor
An invariant to convolution of a signal with an arbitrary function of known maximum size, under assumptions of no noise.
                • Computing similarity between invariants by accounting for their underlying non-Euclidean space (identified with the Grassmann manifold)
                • Applications to single image face recognition under arbitrary uniform and spatially varying blur with a blurred gallery
R. Gopalan, S. Taheri, P. Turaga, and R. Chellappa, "A Blur-robust Descriptor with Applications to Face Recognition", Accepted at IEEE Transactions on Pattern Analysis and Machine Intelligence. [paper]







Face Recognition across Illumination Variations
An integrated study of features insensitive to lighting changes
                • Analysis of performance across different facial regions
                • Classifier combination and applications to single-image face recognition under homogeneous and heterogeneous lighting conditions for gallery and probe

R. Gopalan, and D. Jacobs, "Comparing and combining lighting insensitive approaches for face recognition", Computer Vision and Image Understanding (CVIU) 2010. [paper] [code]



Context-driven Recognition of Objects
Modelling spatial context for object recognition using principles from learning
                • Detecting humans under occlusion by learning semantic context between face and human part detectors using Markov logic networks
                • Localizing lane markings with a pixel-hierarchy contextual descriptor, and learning their relevance with an outlier-robust boosting formulation.
R. Gopalan, and W. Schwartz, "Detecting humans under partial occlusions using Markov logic networks", Performance Metrics in Intelligent Systems Workshop 2010. (Oral, Invited paper) [slides] [paper]
W. Schwartz, R. Gopalan, R. Chellappa, and L.S. Davis, "Robust human detection under occlusion by integrating face and person detectors", International Conference on Biometrics (ICB) 2009. [paper] [code]
R. Gopalan, T. Hong, M. Shneier, and R. Chellappa, "Video-based lane detection using Boosting principles", Snowbird Learning 2009. (Invited paper) [poster] [paper] [code]
R. Gopalan, T. Hong, M. Shneier, and R. Chellappa, "A Learning Approach towards detection and tracking of lane markings", Accepted at IEEE Transactions on Intelligent Transportation Systems [sample tracking results: Seq_1 Seq_2 Seq_3 Seq_4] [results with learning outputs] [paper]



Unsupervised Pattern Discovery with Maximum Margin Principles
Understanding the geometric relation between points and margins by analyzing projections of points on arbitrary line intervals
                • Establishing results that are satisfied only by line segments outside a cluster, for two-cluster and multi-cluster problems
                • Defining a pair-wise similarity measure using these results to perform clustering
R. Gopalan, and J. Sankaranarayanan, "
Max-margin Clustering: Detecting Margins from Projections of Points on Lines
", Accepted at IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2011. [paper] (has minor updates to Prop. 2.5 in the IEEE version) [suppl. material] [poster] [poster-video presentation]








How To Perform Recognition When Data Distribution Changes?
An unsupervised approach that assists the classifier to deal with domain shifts
                • Creating meaningful intermediate domain representations that incrementally represent the effects of change in the marginal.
                • Extensions to semi-supervised adaptation, and multi-domain shift problems.
R. Gopalan, R. Li, and R. Chellappa, "Domain Adaptation for Object Recognition: An Unsupervised Approach
", Accepted at IEEE International Conference on Computer Vision (ICCV) 2011. (Oral) [paper] [partial code]








 
Hand gesture recognition
Shape-driven segmentation and representation for recognizing hand gestures
                • With applications to robotic grasping
R. Gopalan, and B. Dariush, "A vision-based hand gesture interface for robotic grasping", International Conference on Intelligent Robots and Systems (IROS) 2009. (Oral) [paper]
B. Dariush, and R. Gopalan, "
Capturing and recognizing hand postures using inner distance shape contexts", U.S. Patent filed Feb 19 2010, Publication #: 
20100215257
B. Dariush, and R. Gopalan, "Body feature detection and human pose estimation using inner distance shape contexts", U.S. Patent filed 
Feb 19 2010, Publication #: 
20100215271

    

Computationally efficient object localization using contours
An efficient intermediate image representation to encode contour information (under a piece-wise linear approximation of contours)
                • Obtain edge strength of a linear path between arbitrary points in O(1) computations.
                • Applications for fast face/object detection
R. Gopalan, W. Schwartz, R. Chellappa, and A. Srivastava, "Face detection", A Guide to Visual analysis of humans: Looking at people, T. Moeslund et al (Eds), Springer 2011 
[pdf].





 

Sign in|Report Abuse|Print Page|Remove Access|Powered By Google Sites