Yang Bai (白杨)
1414 Circle Drive
Phone: (865)974-6334 (Office)
I'm currently looking for a full-time R&D position in computer vision, digital image processing, machine learning, wireless sensor networks and related. Here is my resume.
digital imaging processing, machine learning, and wireless
visual sensor networks. I'm
extensively involved in projects in terahertz image processing, feature
- based image comparison, and energy
conservation for wireless visual sensor networks.
Extraordinary Professional Promise Award, University of Tennessee, Knoxville, 2010.
Best Thesis Awards, top 5%, Beijing Institute of Technology, 2006.
Renmin Scholarship, top 10%, Beijing Institute of Techology, 2000 - 2004.
Feature-based image comparison algorithm design for wireless visual sensor networks
The Wireless Visual Sensor Network (WVSN) is a group of networked visual sensor nodes with image/video capturing, on board computing and wireless communication capabilities. Although each individual node is build at low cost and its capability is severely limited, working together as a whole, the WVSN can achieve high level computer vision tasks such as environment surveillance, target tracking and object recognition, etc. In order to collaborate with peer nodes, the sensors should form into clusters so that within each cluster all sensors could capture the similar or the same scene. Cluster forming for WVSN should utilize both geographical distance and the Field of View (FOV) information of each visual sensor, because the visual sensor is directional and the visual occlusions are ubiquitous.
Our solution for the
cluster forming relies on the feature-based image comparison algorithm
where each sensor compares the features of the image captured by its
own and its geographical neighbor sensors. Direct image comparison can
discover the visual sensors that are capturing similar or the same
scene, but raw image transmission requires high bandwidth and
consumes a lot energy, both of which are scarce resources for a WVSN.
Feature-based image comparison provides a more compact way for
representing images by using image features, which are more economical
in terms of data transmission. We proposed a light-weight feature point
(corner) detector and found through thorough comparisons that the Scale
Invariant Feature Transform (SIFT) descriptor performs best among all
other feature descriptors, including the Moment Invariant and the
Speeded-Up Robust Feature (SURF) descriptor.
Terahertz Image Processing
Because of their unique capabilities, terahertz waves have demonstrated great potential in the concealed objects detection. However, the limitation in terahertz wave generation/detection devices and the disturbance from the pervasive water molecule absorption have hindered its wide deployment. Terahertz image reconstruction refers to the technique that extract useful information from raw data to produce a meaningful image.
We define the term "differential spectrum" as the difference between the transmitted spectrum and the received spectrum of the terahertz wave, and model the overall absorption effect as a linear combination of the differential spectrum of individual substance. Then a linear unmixing algorithm can be used to separate the overlapped substances represented by their weights, producing several image planes each associated with one of the substances in the image scene.