Parisa Farmanifard

Working on improving the detection of contact lenses in iris images to prevent fraud in iris recognition systems. Deep learning models are used to distinguish between natural irises and those altered by clear or patterned contact lenses, aiming to boost the security of biometric authentication.

A comprehensive benchmark comparing foundation models (VLMs) against domain-specific models for face recognition. Evaluates performance, fusion strategies, and explainability across multiple datasets, revealing tradeoffs between generalizability and specialized accuracy.

Explores the capabilities of GPT-4 multimodal LLM for iris recognition via zero-shot learning across diverse datasets, presentation attacks, occlusions, and real-world variations. A comparative analysis with Gemini Advanced highlights ChatGPT-4's superior performance and robustness in complex iris analysis tasks.

Develops a pixel-level iris segmentation model by fine-tuning the Segment Anything Model (SAM) on ocular images, leveraging Focal Loss to address class imbalance. Achieves 99.58% average segmentation accuracy on ND-IRIS-0405, surpassing the previous best baseline of 89.75%.

Addresses matching faces across different spectral bands (SWIR vs. VIS). Computes ArcFace match scores between synthesized and original cross-spectral images via SG-GAN, achieving well-separated identity histograms and strong discrimination accuracy. Highlights the value of image enhancement for challenging outdoor scenarios.