Feng Tang
[Home] | [Vitae] | [Courses] | [Research] | [Links] | [Contact]
Projects in UCSC:
Tracking as an Online Learning Problem
Object representation using non-orthogonal binary subspaces
Projects before coming to UCSC:
Photograph editing using image inpainting and texture synthesis
Realistic Paintbrush Modeling for a New Painting System
Co-tracking
|
|
This paper treats tracking as a
foreground/background classification problem and proposes an online
semisupervised learning framework. Initialized with a small number of labeled
samples, semi-supervised learning treats each new sample as unlabeled data.
Classification of new data and updating of the classifier are achieved
simultaneously in a co-training framework. The object is represented using
independent features and an online support vector machine (SVM) is built for
each feature. The predictions from different features are fused by combining
the confidence map from each classifier using a classifier weighting method
which creates a final classifier that performs better than any classifier
based on a single feature. The semi-supervised learning approach then uses
the output of the combined confidence map to generate new samples and update
the SVMs online. With this approach, the tracker gains increasing knowledge
of the object and background and continually improves itself over time.
Compared to other discriminative trackers, the online semi-supervised
learning approach improves each individual classifier using the information
from other features, thus leading to a more robust tracker. Experiments show
that this framework performs better than state-of-the-art tracking algorithms
on challenging sequences. Publication: ， Feng Tang Shane Brennan, Qi Zhao and Hai Tao, "Co-Tracking Using Semi-Supervised Support Vector Machines", Proc. ICCV 2007
|
Feature
based tracking (Project Page)
|
|
A good object representation is a description of object/objects by high level features from all perspectives, both spatial and temporal. Extensive representations have been proposed to model objects. They can roughly be categorized as global representation and local representation. Examples of global representations include: color appearance model, subspace methods (like PCA), etc. Local representation describes the object using a set of local features, which are usually selected as the local characterization of object parts. This kind of representation usually incorporates relations between local features to capture the object structure. Typical such representation is region adjacent graph. However, relative less work has been done on modeling the dynamic changes of the object model. Dynamic feature graph is designed as a representation that models both spatial and temporal characteristics of an object. Spatially, the object is represented as an attributed relational graph, with features as nodes and their relations as the edges. Temporally, the graph can adaptively update itself to keep the good features and eliminate unstable features. Publication: ， Feng Tang and Hai Tao, "Probabilistic Object Tracking with Dynamic Attributed Relational Feature Graph", to appear in IEEE Trans. Circuit and Systems for Video Technology, 2008 ， Feng Tang and Hai Tao, "Non-orthogonal Binary Expansion of Garbor Filters with Application in Object Tracking", Proc. of IEEE Motion and Video Computing (WMVC) (oral presentation), 2007 (PDF) ， Feng Tang and Hai Tao, "Object tracking with dynamic feature graphs", in IEEE Workshop on VS-PETS, October 2005. (Oral presentation) ， Feng Tang and Hai Tao, "Dynamic Feature Graph" In CVonline: On-Line Compendium of Computer Vision. R. Fisher (ed). Available: http://homepages.inf.ed.ac.uk/rbf/CVonline |
Object representation using non-orthogonal binary subspaces
Efficient and compact representation of images is a fundamental problem in computer vision. Principal Component Analysis (PCA) has been widely used for image representation and has been successfully applied to many computer vision algorithms. In this paper, we propose a method that uses Haar-like binary box functions to span a subspace which approximates the PCA subspace. The proposed method performs vector dot product using only a few integer additions, therefore dramatically reduces the computational cost. We also show that B-PCA base vectors are nearly orthogonal to each other. As a result, in the non-orthogonal vector decomposition process, the expensive pseudo-inverse projection operator can be approximated by the direct dot product without causing significant distance distortion. Experiments on real image datasets show that B-PCA has comparable performance to PCA in image reconstruction and recognition tasks with speed improvement of two orders of magnitude.
Publications:
- Feng Tang, Ryan Crabb and Hai Tao, "Representing Images Using Non-orthognal Haar-like Bases" IEEE Trans. Pattern Analysis and Machine Intelligence (PAMI) accepted as regular paper, 2007
Feng Tang and Hai Tao, "Fast Multi-scale Template Matching Using Binary Features", Proc. of IEEE Workshop on Applications of Computer Vision (WACV) (oral presentation), 2007
Feng Tang and Hai Tao,"Binary Principal Component Analysis ", Proceedings of British Machine Vision Conference (BMVC) 2006 Volume: 1, On page(s) 377-386, Edinburgh.ISBN: 1-904410-14-6, 2006
Feng Tang and Hai Tao, "Fast Linear Discriminant Analysis Using Binary Bases", IEEE International Conference on Pattern Recognition (ICPR) Volume: 2, On page(s): 52 - 55, Hong Kong. (oral presentation), 2006 and it's journal version in Pattern Recognition Letters (PDF)
Hai Tao, Ryan Crabb, and Feng Tang, "Non-orthogonal binary subspace and its applications in computer vision", Proc. IEEE International Conference on Computer Vision (ICCV) 2005, Beijing, China, 2005
This project is to build a surveillance system that can track multiple objects using stereo cameras.
Feng Tang Mike Harville, Hai Tao and Robinson, Ian N, ``Fusing local appearance features with stereo depth for multi-object tracking", IEEE CVPR workshop on S3D (Search in3D), 2008
Photograph editing using image inpainting and texture synthesis (Project page)
This project presents two methods to repair spoiled photos or remove undesired objects from images.
Texture synthesis is used to fill in the background after removing the undesired objects. Major contributions of our algorithm are: 1) a constraint-based candidate patch searching method which limits the searching within neighboring region with similar texture; 2) a criterion of based on the patch coherence is used to select the best fit candidate which reduces error accumulation and propagation; 3) integration of graphcut optimization to make the seam visually invisible. Experiments show that our system can efficiently handle different cases especially large regions in complex background.
This project presents a novel image inpainting algorithm based on RBF (Radial Basis Functions). After the user selects the regions to be inpainted, the algorithm automatically detects contours of the mask and finds appropriate regions to construct the RBF. Color of the 2D image is treated as height field over a regularly sampled grid, the 2D image inpainting problem is naturally converted to 3D implicit surface reconstruction problem, which RBF has been proved to be a good solver. With RBF resampling£¬the algorithm can nicely fix the damaged image or remove specific objects. Experiments show that our algorithm can fix a large variety of images effectively and robustly.
Recognition: Part of this project won the "Outstanding Prize" of Challenge Cup Research Competition in Zhejiang Province in 2005.
Publication:
， A Novel Texture Synthesis Based Algorithm for Object Removal in Photographs Feng Tang, Yiting Ying, Jin Wang, Qunsheng Peng the 9th Asian Computing Science Conference Chiang Mai, Thailand December,2004, Also M. J. Maher (Ed.): ASIAN 2004, LNCS 3321, pp. 248：C258, 2004.Springer-Verlag Berlin Heidelberg 2004
， Digital Image Inpainting with Radial Basis Functions Tingfang Zhou, Feng Tang, Jin Wang, and Qunsheng Peng, Chinagraph2004 and also Chinese Journal of Image and Graphics
， Hybrid Inpainting and Texture Synthesis for Photograph Editting Feng Tang Technical Report of State Key Lab. of CAD&CG, July.2004
Realistic
Paintbrush Modeling for a New Painting System
(Project page)
|
|
This
project presents a new approach to modeling a 3D physical paintbrush, based
on which an interactive painting system has been developed. Compared with
existent brush-based painting systems, our new system can accurately and
stably simulate the complex painting functionality of a running brush using a
modest amount of system resources. The detailed modeling empowers the user to
create high-quality digital paintings with delicate aesthetic details that
can rival real artwork. With the amount of details to be modelled, we have to
rely on a hierarchical modeling approach, dividing the modeling tasks to
on-line and off-line computations, and a powerful pigment model that is fully
integrated into the brush model. These optimizations and special components
make the system operable in real time, fully interactive, and easy to
manipulate. Recognition: This project won the "Outstanding Prize" of Challenge Cup Research Competition of mainland China in 2003. (Top0.1%), and this is the only "Outstanding Prize" in computer science. Publication:
|
Smooth image
transformation (rotation/scaling)
|
|
Image transformation is a frequently used operation in CAD systems, but how to keep the image smooth in the transformation is a difficult problem, current methods cannot process very well for few color images in textile industry. In this paper we propose a new algorithm for this, first the image is vectorized, using a polygon to approximate the contours of different colors, then we use a dominate point based approach to reduce the number of contour points, then the transformation is applied on the reduced contour points, finally we fill the transformed polygon to reconstruct the final image. Experiments show that this approach can deal with most few color images. Publication: Vectorial Method Based Few-color-image Smooth Transformation Tang Feng, Wang Zhangye, etc (Chinese) NCIG 2003 Volume 8 spec:166-170; and also Chinese Journal of Image and Graphics. |
