IbPRIA 2025: 12th Iberian Conference on Pattern Recognition and Image Analysis

Coimbra, Portugal. June 30 - July 3, 2025

IbPRIA 2025 Accepted Papers

Oral Session 1 - Faces, Body, Fingerprints and Biometrics

A Geometric and Morphometric Methodology for Evaluating Low-Cost 3D Facial Acquisition and Reconstruction Techniques

Álvaro Heredia-Lidón, Alejandro Moñux-Bernal, Alejandro González, Luis M. Echeverry-Quiceno, Mireia Andreu-Montoriol, Susanna Gallardo, Aroa Casado, María Esther Esteban, Neus Martínez-Abadías, Xavier Sevillano

Abstract:
Three-dimensional (3D) facial shape analysis has gained interest due to its potential clinical applications. However, the high cost of advanced 3D facial acquisition systems limits their widespread use, driving the development of low-cost acquisition and reconstruction methods. This study introduces a novel evaluation methodology that goes beyond traditional geometry-based benchmarks by integrating morphometric shape analysis techniques, providing a statistical framework for assessing facial morphology preservation. As a case study, we compare smartphone-based 3D scans with state-of-the-art deep learning reconstruction methods from 2D images, using high-end stereophotogrammetry models as ground truth. This methodology enables a quantitative assessment of global and local shape differences, offering a biologically meaningful validation approach for low-cost 3D facial acquisition and reconstruction techniques.

Abnormal Human Behaviour Detection using Normalising Flows and Attention Mechanisms

Ana Filipa Rodrigues Nogueira, Hélder P. Oliveira, Luís F. Teixeira

Abstract:
The aim of this work is to explore normalising flows to detect anomalous behaviours which is an essential task mainly for surveillance systems-related applications. To accomplish that a series of ablation studies varying the parameters of the STG-NF model [3] and combining it with attention mechanisms were performed. Out of all these experiments, it was only possible to improve the state-of-the-art result for the UBnormal dataset by 3.4 percentual points (pp), for the Avenue by 4.7 pp and for the Avenue-HR by 3.2 pp. However, further research remains urgent to find a model that can give the best performance in the different scenarios. The inaccuracies of the pose tracking and estimation algorithm seems to be the main factor limiting the models' performance.

Federated Learning for Secure and Privacy-Preserving Facial Recognition: Advances, Challenges, and Research Directions

Ajnas Muhammed, João Marcos, Nuno Gonçalves

Abstract:
Federated learning is an innovative, decentralized machine learning paradigm that allows multiple devices or entities to collaboratively train a shared model without transferring data to a central server. By keeping data localized, this distributed approach ensures enhanced privacy and security for each participating node. Facial recognition, a rapidly evolving field, leverages deep learning techniques to achieve remarkable advancements, often surpassing human-level performance on certain datasets. However, the sensitive nature of facial data, which contains personally identifiable information, raises significant privacy and security concerns. Federated learning has emerged as a promising solution to address these privacy challenges in the facial recognition community. This paper presents a comprehensive review of existing literature on facial recognition frameworks utilizing federated learning. The reviewed techniques are systematically categorized to provide a structured analysis, emphasizing their contributions and relevance to the broader domain of federated learning-based facial recognition. Specifically, this work aims to summarize and analyze various federated learning-based facial recognition methods, their underlying techniques, and their objectives. Furthermore, it offers a high-level perspective on how different functionalities and design principles of federated learning have been applied in facial recognition applications. By doing so, this review identifies key challenges and highlights promising research directions for future advancements in the field.

Multi-scale Temporal Pose analysis for Gait Recognition

Nicolás Cubero, Francisco M. Castro, Nicolás Guil, Manuel J. Marín-Jiménez

Abstract:
The problem of gait recognition has primarily focused on using silhouette or visual modalities to describe the gait cycle. While these methods offer rich representations, they are heavily influenced by visual covariates like body contours or carrying objects. Pose-based methods provide greater robustness against these covariates, but current approaches have yet to effectively extract rich features from pose sequences, leading to suboptimal performance. In this work, we introduce MuSTGaitPose, a model architecture that implements multi-scale temporal analysis of pose sequences to extract richer gait features. Our model features the Multi-scale Temporal Block (MuST Block), which scans pose sequences at multiple time scales to identify key temporal patterns at each scale. We also developed Multi-scale Temporal Attention Fusion (MuSTAF) to optimally aggregate the multi-scale features based on their relative importance at each spatial and temporal location. Thus, our approach produces a combined feature that emphasizes the most relevant gait patterns across all gridded time scales. Additionally, we leverage pose heatmaps for a richer descriptor. Extensive experiments show that our approach outperforms previous pose-based methods, achieving mean Rank-1 accuracies of 90.9% on the CASIA-B and 86.2% on the SUSTech1K datasets, as well as a true acceptance rate of 95.8% at a false acceptance rate of 1% on the FVG-B dataset.

Pseudo-MOS Learning: A Hybrid Full-to-No-Reference FIQA Framework

André Neto, Nuno Gonçalves

Abstract:
A persistent discrepancy exists between standard Image Quality Assessment (IQA) metrics and human perceptual judgments, typically quantified through Mean Opinion Scores (MOS). This gap poses a critical challenge for tasks where visual quality directly impacts performance, such as facial recognition and visual data transmission. In particular, assessing the perceptual quality of steganographically distorted facial images remains difficult, especially in the absence of reference images. To address this, we introduce a hybrid Full-to-No-Reference framework for Face Image Quality Assessment (FIQA), built upon a learning strategy based on pseudo-MOS. A full-reference fusion metric is first trained by regressing multiple classical IQA scores against human MOS on a subset of a facial dataset. This metric is then applied to the full dataset to generate pseudo-MOS labels. Using deep features extracted from a ResNet-18 model pretrained on ImageNet, we train a no-reference regressor capable of predicting perceptual quality. The proposed framework bridges full-reference supervision and no-reference inference, offering a scalable and accurate solution for FIQA in challenging conditions and paving the way for application-specific, data-driven IQA designs.

Writer Identification using Simplified Handwritten Text Recognition Models

Alejandro H. Toselli, Álvaro Cuéllar, Sònia Boadas, Enrique Vidal, Joan Andreu Sánchez

Abstract:
The theater of the Spanish Early Modern period is a textual collection composed of thousands of works and hundreds of playwrights and is one of the greatest examples of Spanish and Western literature. These works often present a wide range of textual problems stemming mainly from the copying process itself given that it may alter the meaning of the original works by the authors. Therefore, identify autograph testimonies written directly by the playwrights themselves, are of particular importance. To perform this identification automatically, we address an approach for writer identification that combines Deep Convolutional and Recurrent Neural Networks with n-gram language models. This is posed as a classification problem, introducing a probabilistic framework that goes beyond the plain classification of recognized handwritten text. Preliminary experiments are conducted to validate our proposal to distinguish between Lope de Vega's manuscripts and other non-Lope hands.

Publisher

Endorsed by

Technical Sponsors