Jia-Bin Huang

  • Advisor:
      • Narendra Ahuja
  • Departments:
  • Areas of Expertise:
      • Artificial intelligence
      • Computational Photography
      • Machine Learning
      • Computer Vision
  • Thesis Title:
      • Physically Grounded Visual Constraints for Inverse Problems in Image and Video Processing
  • Thesis abstract:
      • Inverse problems arise from various image and video processing tasks, including image denoising, super-resolution, deblurring and completion. Due to the illposed nature, effective regularization terms are the key elements for addressing these inverse problems. Over the last decades, there have been rapid progress on learning natural image priors and applying them to various problems. However, existing work often exploit such priors by treating an image as a pure 2D signals, ignoring the fact that an image is a projection of the 3D physical world. In this thesis, we show that by incorporating scene-specific geometric constraints we are able to achieve the-state-of-the-art performance in image completion, image super-resolution and video completion problems. This suggests that these physically grounded visual constraints can dramatically improve existing algorithms for solving inverse problems in computer vision.
  • Downloads:

Contact information: