Computer Vision and Graphics (RMSE - not root mean squared error): 

  • Recognition: How to learn a model from large-scale high-dimensional visual data for object or human action classification? 

  • Mining: How to efficiently find model samples from large-scale multidimensional data in terms of memory and computation?

  • Synthesis: How to generate novel samples from learned model? How to implement efficient rendering, such as scientific visualization and image-based rendering, using learned models of large-scale data in a personal computer?

  • Estimation: How to estimate a model with physical constraints, for example, estimate geometric and photometric models, such as 3D shape, reflectance and lighting models  of real-world 3D scenes?

Other Interests:

  • Medical image processing: X-ray and CT image restoration and enhancement, CT and Ultrasound multi-image compounding, etc.

  • Computational photography/videography: image/video editing, HDR image/video, etc.

  • Time series analysis: system monitoring, gene expression sequence analysis, etc.

  • CAD and geometrical modeling


Recognition: Multilinear Models
  • Dimensionality reduction using Datum-as-Is representation (ICPR'04, CVPR'05)
  • Face and facial expression recognition (ICCV'03)


Recognition: Probabilistic Graphical Models
  • Learning human motion categories
  • pLSA and LDA for human action recognition
  • Motion localization in long and complex video sequences (BMVC'06, VPCVPR'06)


Mining: Find Model Samples Efficiently
  • Incremental tensor subspace learning
  • Efficient video retrieval for large-scale data
  • Still and dynamic texture synthesis


  Synthesis: Data-Driven Models for Large-Scale Image-Based Rendering
  • Out-of-core tensor approximation (SIGGRAPH'05)
    • Bidirectional texture function (BTF- 6D)
    • Dynamic BTF(DBTF - 7D)
    • Temporal volumn data (4D)
  • Hierarchical Tensor Approximation (TCGA)


Estimation: Geometric and Photometric Estimation and Editing
  • 3D Shape, BRDFs, and illumination estimation
  • Relighting with estimation (EGSR'06, SIGGRAPH'06 Sketch)




Medical Image Processing
  • Multi-Image Compounding: Ultrasound/CT Images Frequency Compounding (ICASSP'05, ICCV'05)
  • Image Enhancement and Restoration (ICIP'06)

(work at Siemens Corporate Research)

Computational Videography

(work at Mitsubishi Electric Research Lab)

  Time Series Analysis: A Statistical Method (under reconstruction)
  • Gene Expression Sequence Analysis
  • Computer System Event and Change Detection

(work at IBM Almaden Research Center)

  Geometry Modeling for CAD/CAM
  • Process-Conscious Toolpath Simulation and Optimization Using Quintic Spline Modeling (IMECE'02, JMSE'05)

Updated: Jan.1, 2006