Building text features for object image classification

We introduce a text-based image feature and demon- strate that it consistently improves performance on hard object classification problems. The feature is built using an auxiliary dataset of images annotated with tags, down- loaded from the internet. We do not inspect or correct the tags and expect that they are noisy. We obtain the text fea- ture of an unannotated image from the tags of its k-nearest neighbors in this auxiliary collection. A visual classifier presented with an object viewed un- der novel circumstances (say, a new viewing direction) must rely on its visual examples. Our text feature may not change, because the auxiliary dataset likely contains a similar pic- ture. While the tags associated with images are noisy, they are more stable when appearance changes. We test the performance of this feature using PAS- CAL VOC 2006 and 2007 datasets. Our feature performs well, consistently improves the performance of visual ob- ject classifiers, and is particularl...

• Type: Conference
• Year: 2009
• Where: CVPR
• Authors: David A. Forsyth, Derek Hoiem, Gang Wang
• Keywords: Computer Vision, CVPR 2009, Image Classification