INESC TEC - Institute for Systems and Computer Engineering, Technology and Science (INESC TEC)

Funder

Organizational Unit

Publications

A low resource skeleton maturation estimation system for automatic hand X-Ray assessment in pediatric applications

Publication . Campos, Ana; Silva, Maria; Azeredo, Ricardo; Coelho, Luis; Reis, Sara; Abreu, Sílvia

The assessment of differences between skeletal age and chronological age in childhood is often based on the comparison of the patient's left hand x-ray with a reference atlas, performed by a experienced professional. This procedure involves a manual image analysis, that can be subject to inter rater variability posing several problems for clinical applications. In this paper a new methodology for skeleton maturation estimation based on automatic hand X-ray assessment for pediatric applications on a low resource devices (e.g. mobile device) is proposed. The pipeline covers hand-area estimation and bone-area estimation to achieve maturation scores which are then indexed with references images, separately for male and female. The proposed approach is based on simple image processing functions always bearing in mind the application on a mobile context. The involved steps are thoroughly presented and all the used functions are explained. The performance of the system was then evaluated using the complete pipeline. The obtained results pointed to an average error rate of 15,38±3,31%, which is subject to improvements. In particular, contrast enhancement in some lower quality images still offers some challenges.

2023Conference object

Restricted access

Benchmarking computer-vision-based facial emotion classification algorithms while wearing surgical masks

Publication . Coelho, Luis; Reis, Sara; Moreira, Cristina; Cardoso, Helena; Sequeira, Miguela; Coelho, Raquel

Effective human communication relies heavily on emotions, making them a crucial aspect of interaction. As technology progresses, the desire for machines to exhibit more human-like characteristics, including emotion recognition, grows. DeepFace has emerged as a widely adopted library for facial emotion recognition. However, the widespread use of surgical masks after the COVID-19 pandemic presents a considerable obstacle to its performance. To assess this issue, we conducted a benchmark using the FER2013 dataset. The results revealed a substantial performance decline when individuals wore surgical masks. “Disgust” suffers a 22.6% F1-score reduction, while “Surprise” is least affected with a 48.7% reduction. Addressing these issues improves human–machine interfaces and paves the way for more natural machine communication.

2023Journal article

Open access