- Computer Vision
- Image/Video Processing
- Machine Learning
|
- [C1] Chen Cao, Shifeng Chen, Yuhong Li, and Jianzhuang Liu,
" Online Non-feedback Image Re-ranking via Dominant Data Selection ", ACM Multimedia (ACM MM '12), Nara, Japan, 2012. [Paper]
- [C2] Chen Cao, Shifeng Chen, Changqing Zou, and
Jianzhuang Liu, " Locating High-density Clusters with Noisy Queries ", International Conference on Pattern Recognition (ICPR '12), Tsukuba, Japan, 2012. [Paper]
- [C3] Chen Cao, Shifeng Chen, Wei Zhang, and
Xiaoou Tang, " Automatic Motion-Guided Video Stylization and Personalization ", ACM Multimedia (ACM MM '11), Scottsdale, AZ, USA, 2011. [Paper]
|
Journal Paper
- [J1] Wei Zhang, Chen Cao, Shifeng Chen, Jianzhuang Liu and Xiaoou Tang, " Style Transfer via Image Component Analysis ", IEEE Transactions on Multimedia (T-MM '13), 2013. [Paper] [DemoPage] [DemoVideo]
|
Patent
- [P1] Shifeng Chen, Chen Cao, Wei Zhang, Jianzhuang Liu, Xiaoou Tang, Yu Qiao, " Video Style Transfer Method and System " China Patent NO.201110283909.9, Filed on Semptember 2011, Issued on February 2012.
|
|
Experiences
- Master in Computer Engineering, University of Florida, Gainesville, FL, 2012.8 - present
- Research Associate, Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, Shenzhen, China, 2011.7 -
2012.7
- Bachelor in Electrical Engineering, University of Science and Technology of China, Hefei, China, 2007.9 - 2011.7
|
Projects
- Example-based image rendering via image component analysis. Given an arbitrary source image and a style template image, both images will be decomposed into three components: draft, paint, and edge. Then style transfer is conducted on paint and edge components from style template to source image through Markov Random Fields minimization process. For more examples and high resolution images please refer to http://mmlab.siat.ac.cn/personal/style/supply/SupplementalMaterial/results.html. This work is published in IEEE Transactions on Multimedia (T-MM), 2013. Please refer to my [Paper] for details.
 |
 |
 |
| Source Image |
"Craquelure" Style |
“Oil-painting” Style |
 |
 |
 |
| “Pastel” Style |
"Oil-Canvas" Style |
"Watercolor" Style |
| |
| |
- Video stylization and personalization. Video stylization transfers a source video into art styles. The technique not only uses non-photorealistic rendering methods to stylize each frame respectively, but considers the temporal coherence between frames via optical flow. Video personalization detects and stylizes human faces in video specifically to keep these regions distinguishable. Here is an example for watercolor transfer [Video]. This work is published in ACM Multimedia (MM), 2011. Please refer to my [Paper] for details.
 |
 |
 |
| The 1st frame |
"Watercolor" template |
Watercolor-like frame |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
| |
| |
- Online non-feedback Image re-ranking. This work aims at improving the precision of keyword-based image search engine. Our method builds spectral graph on image visual features to explore data's intrinsic distribution. We have observed that data in dominant clusters are more likely to be the search target, and isolated data are noise. From this view we devise image re-ranking algorithm that high-accurate, automatic and real-time. The work is published in ACM Multimedia (MM), 2012. Please refer to my [Paper]
for details.
- High density cluster learning. In several real world applications, the objective data are distributed in high density clusters. Based on the perspective of spectral clustering, we propose a simply but effective learning method to explore the high density clusters in data distribution automatically, and implement the algorithm to web image re-ranking and image co-segmentation. This work is published in International Conference on Pattern Recognition (ICPR), 2012. Please refer to my [Paper] for details.
 |
|
 |
| Web search. (1) Search engine. (2) Re-ranking results |
|
Co-segmentation. (Left) Image pairs. (Right) Segments |
| |
|
|
