Share on Facebook Tweet on Twitter Share on LinkedIn Share by email
Computer Vision

Teaching computers to understand the visual world

The goal of computer vision is to make computers efficiently perceive, process, and understand visual data such as images and videos. The ultimate goal is for computers to emulate the striking perceptual capability of human eyes and brains-or even to surpass and assist the human in certain ways.

Within Microsoft Research, our computer-vision research include investigations into:

  • Imaging and Photogrammetry, including high-resolution cameras, radiometric calibration, photometric stereo, 3-D imaging and video, 3-D scene reconstruction from images and video, and image and video enhancement.
  • Pattern Recognition and Statistical Learning, including data clustering and classification, manifold learning, and high-dimensional geometry and statistics.
  • Object Detection and Recognition, including face detection, alignment, and tagging; video-based face recognition; and sparsity-based robust face recognition. We also investigate general object-class recognition and advanced medical-image analysis.
  • Image and Video Editing and Enhancement, including denoising and deblurring, novel representations for images and video, techniques for content-aware edits such as in-painting, and object removal. 

J. Margeta, A.Criminisi, D.C.Lee, and N.Ayache, Recognizing Cardiac Magnetic Resonance Acquisition Planes using Finetuned Convolutional Neural Networks, in To appear in Computer Methods in Biomechanics and Biomedical Engineering, December 2015.

H. Lombaert, A. Criminisi, and N. Ayache, Spectral Forests: Learning of Surface Data, Application to Cortical Parcellation, in Medical Image Computing and Computer Assisted Intervention (MICCAI), Springer, October 2015.

Gerard Pons-Moll, Jonathan Taylor, Jamie Shotton, Aaron Hertzmann, and Andrew Fitzgibbon, Metric Regression Forests for Correspondence Estimation, in IJCV, Springer, August 2015.

Wangjiang Zhu, Baoyuan Wang, and Steve Lin, Adaptive Pooling over Multiple Trajectory Attributes for Action Recognition, in AVSS 2015, 12th IEEE International Conference On Advanced Video and Signal-based Survillance, August 2015.

More publications ...