Research

Research at CILab

CILab conducts research at the intersection of image processing, computer vision, computational imaging, and machine learning. We develop interpretable and reliable algorithms by integrating physical imaging models, optimization principles, and data-driven learning. Our current research focuses on the following topics.

Model-Based Deep Learning

• We develop interpretable deep networks by translating iterative optimization algorithms into trainable architectures. We solve challenging inverse imaging problems by combining physical imaging models, optimization structures, and learned priors, which can improve data efficiency and generalization.

• Our research includes:

  • Deep unfolding for inverse imaging problems

  • Low-rank matrix and tensor recovery

  • Physics-driven and prior-guided network design

  • Task-guided model-based deep learning

Computational Imaging

• We develop computational imaging systems that jointly optimize image acquisition, reconstruction, and subsequent computer vision tasks. Our research combines imaging models with neural processing to recover high-quality images from incomplete, degraded, or nonconventional sensor measurements.

• Our research includes:

  • Neural image signal processing

  • RGBW remosaicing and color reconstruction

  • Single-shot and multi-exposure high dynamic range imaging

Image Restoration and Enhancement

• We develop robust algorithms for recovering high-quality images and videos degraded by adverse imaging conditions, sensor limitations, environmental interference, and display-camera interactions. We explore cutting-edge learning approaches, e.g., contrastive learning, transformer-based architectures, and generative models.

• Our research includes:

  • Underwater image enhancement

  • Image and video deblurring

  • Screenshot demoiréing

  • Low-light and contrast enhancement

Generative Restoration and Semantic Visual Communication

• We develop controlled generative algorithms for restoring and communicating visual information under extrem compression, severe degradation, and incomplete observations. We focus on guiding generative priors using measurements, semantic information, and task-specific constraints.

• Our research includes:

  • Diffusion-based image and video restoration

  • Text- and image-conditioned multimodal restoration

  • Restoration of extremely compressed images and videos

  • Generative restoration for video coding for machines

Multimodal, Spectral, and Remote Sensing Imaging

• We develop image fusion and reconstruction algorithms that exploit complementary information from multiple sensors, spectral bands, and imaging modalities by modeling cross-modal correlations and structural consistency.

• Our research includes:

  • Infrared and visible image fusion

  • Multispectral and hyperspectral image restoration

  • Pansharpening

  • SAR-guided cloud removal in hyperspectral imagery