10/16/2009 Forrest Iandola
Written by Forrest Iandola
A paper co-authored by CS PhD student Hossein Mobahi, ECE PhD student Shankar R. Rao, and ECE alumnus Allen Y. Yang earned the San Uk Lee Best Paper Award at the Asian Conference on Computer Vision (ACCV 2009).
In the winning paper, entitled “Natural Image Segmentation with Adaptive Texture and Boundary Encoding,” the authors propose a novel method for natural image segmentation. The segmentation method is based on an algorithm called Texture and Boundary Encoding Segmentation (TBES), which encodes image texture and boundary information, “using a Gaussian distribution and adaptive chain code.”
The authors found it challenging to evaluate the effectiveness of TBES because, “there is little consensus on what criteria should be used to evaluate the quality of image segmentations.” Even so, the authors used two metrics, Probabilistic Rand Index (PRI) and Variation of Information (VOI) to test their algorithm’s ability to compress images with minimal loss. The PRI and VOI data from the novel TBES segmentation algorithm compares favorably with five commonly used image compression algorithms.
The winning paper was presented at ACCV, which was held September 23-27 2009 in Xi’an China. ACCV attracts computer vision researchers and developers from around the world. Presentation topics include super-resolution, color and texture, and face and gesture analysis. The conference is sponsored by Microsoft, Fujitsu, and the National Natural Science Foundation of China.