7 February 2011 Learning object detectors from online image search
Author Affiliations +
Abstract
Being able to detect distinguishable objects is a key component in many high level computer vision applications. Traditional methods for building such detectors require a large amount of carefully collected and cleaned data. For example to build a face detector, a large number of face images need to be collected and faces in each image need to be cropped and aligned as the data for training. This process is tedious and error-pruning. Recently more and more people are sharing their photos on the internet, if we could leverage these data for building a detector, it will save tremendous amount of effort in collecting training data. Popular internet search engines and community photo websites like Google image search, Picassa, Flickr make it possible to harvesting online images for image understanding tasks. In this paper, we develop a method leveraging images obtained from online image search to build an object detector. The proposed method can automatically identify the most distinguishable features across the downloaded images. Using these learned features, a detector can be built to detect the object in a new image. Experiments show promising results of our approach.
© (2011) COPYRIGHT Society of Photo-Optical Instrumentation Engineers (SPIE). Downloading of the abstract is permitted for personal use only.
Feng Tang, Daniel R. Tretter, "Learning object detectors from online image search", Proc. SPIE 7879, Imaging and Printing in a Web 2.0 World II, 78790N (7 February 2011); doi: 10.1117/12.876685; https://doi.org/10.1117/12.876685
PROCEEDINGS
8 PAGES


SHARE
Back to Top