da/sec scientific talk on Biometrics

Topic: Not My Hands: Identity-guided Diffusion for Generating Synthetic Hand Images

by Robert Nichols
D19/2.03a, March 28, 2024 (Thursday), 12.00 noon

Keywords — Hand recognition, Forensic analysis, Synthetic Image Generation, Deep learning, Diffusion models

Abstract

„Recognition of individuals based on images of their hands does not enjoy the same level of publicity as e.g. face recognition and therefore work on this topic is limited. Nevertheless, the utility in specific scenarios is considerable; for instance in forensic contexts, where investigators often face the difficult task of identifying a suspect with little more than an image depicting only parts of a hand i.e. the palmar or dorsal aspect of hand.
However, the required large amounts of data required for robust recognition systems is frequently unavailable. Recent advancements in the area of generative artificial intelligence have demonstrated impressive capabilities in terms of image fidelity and performance, in turn enabling the substitution of real images with synthetically generated samples.
To this end, we propose a method to generate mated hand images by leveraging prior work on Latent Diffusion Models (LDM) and state-of-the-art hand recognition to extract meaningful feature embeddings to condition the diffusion model. Following this process, we create a new dataset of synthetic hand images and assess the ability to generate mated samples as well as the distinctiveness of the synthetic subjects.“