April 14, 2010 : Ce Liu

Microsoft Research New England

Exploring new models and features for visual recognition

In this talk, I will show our strategies to visual recognition, using object recognition and material recognition as examples. First, we introduce a new pipeline for object recognition, label transfer, which transfers object labels from human annotation of training images to parse an input image. We use SIFT flow, a technique to align images across 3D scenes, to reliably match an input image to the images in our database. Experimental results show that, although no classifier is used, our nonparametric scene parsing system can be a winner over traditional training-based systems.

Unlike other visual recognition tasks, it is difficult to find good, reliable features that can tell material categories apart. Our strategy is to use a rich set of low and mid-level features that capture various aspects of material appearance. We propose an augmented Latent Dirichlet Allocation (aLDA) model to combine these features under a Bayesian generative framework and learn an optimal combination of features. Experimental results show that our system performs material recognition reasonably well on a challenging material database, outperforming state-of-the-art material/texture recognition systems.

seminars/seminaritems/2010-04-14.txt · Last modified: 2010/04/07 14:36 by koray