|

Intelligent Computing. CompCom 2025. Lecture Notes in Networks and Systems, vol 1425. Springer, Cham

BioCube is a hybrid encryption framework that enhances the security of finger-vein biometrics. Biometric templates are transformed into secure, image-based representations using two techniques: a 2D barcode encoder (JAB code) and a pseudo-random probabilistic (PRP) generator. These images are further processed with transformations inspired by the infinity cube puzzle, producing highly secure templates while ensuring performance, unlinkability, reusability, and non-invertibility.

SN Computer Science Volume 6, article number 17, (2025)

This work focuses on defending deep neural networks against attacks on image recognition systems where small, carefully placed patches can trick the system into making wrong predictions. The method automatically detects these patches and fills in the missing or altered parts of the image, restoring it to its original form. It works in black-box settings without prior knowledge of the patch’s location, shape, or size, and does not require retraining the network. This makes it practical for real-world applications like autonomous vehicles and surveillance systems, ensuring reliable and secure image classification.

Computer Analysis of Images and Patterns. CAIP 2023. Lecture Notes in Computer Science, vol 14184. Springer, Cham.

This project tackles attacks on image recognition systems where small, malicious patches can cause the system to misidentify images. The method automatically detects these patches and restores the affected areas using advanced image inpainting techniques, preserving the original structure of the image. It works without prior knowledge of the patch’s location, shape, or size, and does not require retraining the system, making it practical for real-world applications. The approach improves system reliability and ensures accurate classification even under attack.