Heterogeneous Face Recognition Using Inter-Session Variability Modelling

Abstract

The task of Heterogeneous Face Recognition consists in to match face images that were sensed in different modalities, such as sketches to photographs, thermal images to photographs or near infrared to photographs. In this preliminary work we introduce a novel and generic approach based on Inter-session Variability Modelling to handle this task. The experimental evaluation conducted with two different image modalities showed an average rank-1 identification rates of 96.93% and 72.39% for the CUHK-CUFS (Sketches) and CASIA NIR-VIS 2.0 (Near infra-red) respectively. This work is totally reproducible and all the source code for this approach is made publicly available.

Cross-eyed 2017: Cross-spectral iris/periocular recognition competition

“…The Universe Background Model (U BM ) is modelled with 512 Gaussians and the dimension of the session variability matrix (U ) is 160. Implementation details of this input can be found in [25]. The Geodesic Flow Kernel (GFK) models the source domain and the target domain with d-dimensional linear subspaces and embeds them onto a Grassmann manifold.…”

Prepended Domain Transformer: Heterogeneous Face Recognition Without Bells and Whistles

“…It can be seen that the proposed approach achieves an average Rank-1 accuracy of 97.1% with a standard deviation of 1.3%, which is much greater than the results from other baselines reported in the literature. [42] 78.72 % (-) Base-PLS in [42] 53.05% (-) ines LBPs + DoG features in [18] 36.8% (3.5) Repro-ISV in [51] 23.5% (1.1) ducible GFK in [52] 34.1% (2.9) baselines DSU(Best Result) [13] 76.3% (2.1) Reproducible DSU-Iresnet100 88.2% (5.8) Reproducible PDT (Proposed) 97.1% (1.3) Reproducible…”

Quantifying the impact of image degradation on Deep Learning models in face recognition systems

“…A prominent example of the success of the facial recognition dataset is the Labeled Faces in the Wild (LFW) [Huang et al 2007], widely used as a performance benchmark for such applications. Many algorithms now achieve accuracy scores close to 100%, indicating the relative ease of recognizing faces in this dataset [de Freitas Pereira et al 2022]. However, despite the achievements in many application domains, facial recognition encounters challenges in uncontrolled environments where image capture conditions are adverse.…”

“…However, despite the achievements in many application domains, facial recognition encounters challenges in uncontrolled environments where image capture conditions are adverse. In applications such as autonomous cars, public surveillance, low-light areas, or low-quality capture equipment, external images often fail to meet the ideal criteria for accurate processing [de Freitas Pereira et al 2022, Grm et al 2018. Numerous studies have explored noncontrolled environments, and the results consistently indicate lower scores than controlled environments [Schlett et al 2022].…”