Iris Image Synthesis (funded by WVU)
- Generated high-quality synthetic iris samples with unique identities and seven distinct anomalies using Gen-AI (StyleGAN and Diffusion Model), ensuring no overlap with training data.
- Developed a model fore enhancing and degrading the visual quality of synthetic iris images and modifying physical features of the iris.
- Improved the fidelity of iris recognition and presentation attack detection models using synthetic samples, while preserving individual identity.
- Conducted and analyzed perceived identity match through a 263-subject human study via Qualtrics and Prolific.
Gender Bias in Iris Presentation Attack (funded by WVU)
- Evaluated the performance of deep learning iris presentation attack detection (PAD) across three datasets using six different methods (Deep Learning and non-Deep Learning).
- Analyzed gender-based differences in iris PAD susceptibility by comparing detection effectiveness for male and female irises.
- Identified and mitigated bias in iris PAD algorithms by examining image differences that affect detection accuracy.
Detection and Localization of Forgeries in CT Scan Images
- Trained and evaluated the Single Shot MultiBox Detector (SSD) for detecting and localizing image forgery in CT scans.
- Tested capacity of a natural image tampering detection on medical images.
Human Emotional Responses to Images Generated by Deep Dream
- Analyzed and visualized how visual image properties influence human emotional responses, measured through arousal and valence dimensions.
RV Trip Pattern (funded by Lippert Components Team)
- Explored, analyzed and visualized raw RV data in SQL to identify travel patterns before and after COVID.
- Provided insights into how the pandemic affected RV usage.
- Recommended areas for improvement and investment based on the findings.
General-Info 