Fan Wang: 2014 Networks of Shapes, Images, and Programs: The Power of Joint Data Analysis Workshop

 

Wednesday, April 16, 2014
Location: Fisher Conference Center, Arrillaga Alumni Center

"Network-Based Image and Shape Co-Segmentation"
11:00am - 11:30am

Abstract:

Co-segmentation of image/shape sets has great importance for object recognition, classification, and retrieval. In this talk, we will first introduce a co-segmentation technique by putting images in a network context and jointly segmenting them with almost no supervision. To allow the images to share segmentation information with each other, we build a network that contains all images, and extract functional maps between connected image pairs based on image appearance features. These functional maps act as general property transporters between the images and are used to transfer segmentation functions. The functional maps are also regularized to approximately satisfy cycle-consistency under composition in the network. A joint optimization framework is then proposed to simultaneously generate all segmentation functions over the images so that they both align with local segmentation cues in each particular image, and agree with each other under network transportation. This collective effect of the joint processing using functional maps yields superior segmentation results on image collections, and then is naturally extended to co-segment shape collections. Extensive experiments have demonstrated the improvement of our technique over the state-of-the-art methods in both image and shape co-segmentation.


Bio:

Fan Wang is a PhD candidate in Department of Electrical Engineering in Stanford University. She obtained her Bachelor's and Master's degree majored in Control Theory both from Tsinghua University. She is generally interested in computer vision, machine learning, and optimization. Her current research is mainly focused on joint understanding of image/shape collections.