Image Watermarking with Deep Learning

Watermarking refers to the covert embedding of information into a host/cover medium. In this work, we develop deep learning-based watermarking methods that achieve adaptivity and robustness through specialized training strategies and architectural design. A key application is reliable watermark extraction from camera-resampled images, enabling end users to recover embedded information by simply scanning a captured image. 

Funded: NSF CRII 2104267

Selected Publications:

• [1] TIACam: Auto-augmentation and text-anchored invariant feature learning for camera robustness [https://arxiv.org/abs/2602.18863];

• [2] Survey of Deep Learning-Based Image Watermarking [https://www.mdpi.com/2076-3417/13/21/11852];

• [3] Connecting Tradition and Deep Learning Image Watermarking [https://arxiv.org/abs/2007.02460];

Multimodal Representation Learning for Phenotype Analysis

This project, conducted in collaboration with a biology research team, develops multimodal learning methods to model behavioral and biological patterns in marmosets. We integrate heterogeneous data sources, including video, audio, and microbiome measurements, to learn unified latent representations that capture both behavioral and biological features. These representations enable the analysis of temporal phenotype trajectories and the discovery of bio-behavioral clusters, providing a data-driven framework for understanding complex biological systems. 

Collaborators: UNO Biology;  UNO Biomechanics

Hazards Precursor Prediction

This project is part of a multi-team effort to enhance community resilience to extreme wind hazards. Our work develops cross-modal self-supervised learning methods to learn robust representations of meteorological patterns from both historical and real-time data. Historical data captures stable, long-term dynamics, while real-time data enables continuous adaptation to evolving conditions, supporting reliable precursor prediction. 

Funded: NSF 2431053

Collaborator: UNO Aviation Institute

Selected Publications:

• Interpretable Deep Learning for Wind Hazard Prediction [https://arxiv.org/abs/2505.14522];