An Additive Latent Feature Model for Transparent Object Recognition

TitleAn Additive Latent Feature Model for Transparent Object Recognition
Publication TypeConference Paper
Year of Publication2009
AuthorsFritz, M., Black M., Bradski G., Karayev S., & Darrell T.
Other Numbers3175
Abstract

Existing methods for visual recognition based on quantized local features can performpoorly when local features exist on transparent surfaces, such as glass orplastic objects. There are characteristic patterns to the local appearance of transparentobjects, but they may not be well captured by distances to individual examplesor by a local pattern codebook obtained by vector quantization. The appearanceof a transparent patch is determined in part by the refraction of a backgroundpattern through a transparent medium: the energy from the background usuallydominates the patch appearance. We model transparent local patch appearanceusing an additive model of latent factors: background factors due to scene content,and factors which capture a local edge energy distribution characteristic ofthe refraction. We implement our method using a novel LDA-SIFT formulationwhich performs LDA prior to any vector quantization step; we discover latent topicswhich are characteristic of particular transparent patches and quantize the SIFTspace into transparent visual words according to the latent topic dimensions. Noknowledge of the background scene is required at test time; we show examplesrecognizing transparent glasses in a domestic environment.

Acknowledgment

This work was supported in part byWillow Garage, Google, NSF grants IIS-0905647 and IIS-0819984, and a Feodor Lynen Fellowship granted by the Alexander von HumboldtFoundation.

URLhttp://www.icsi.berkeley.edu/pubs/vision/additivelatent09.pdf
Bibliographic Notes

Proceedings of the 23rd Annual Conference on Neural Information Processing Systems (NIPS 2009), Vancouver, British Columbia, Canada

Abbreviated Authors

M. Fritz, M. Black, G. Bradski, S. Karayev, and T. Darrell

ICSI Research Group

Vision

ICSI Publication Type

Article in conference proceedings