Abstract

Despite Handwritten Text Recognition (HTR) has been addressed for decades, the recognition of images of manuscripts is challenging. Current deep-learning architectures are powerful but they need a huge amount of training data. Not surprisingly, deep learning methods are not satisfactory in case of low resource scenarios (when data or labeled data is scarce). This is the case of historical manuscripts, music scores, ciphered manuscripts or documents with rare scripts. This talk will discuss some of the current challenges and overview some methods designed for coping with the limitations of few labelled data, including few-shot learning, synthetic data generation and self-supervised learning.

Biography

Dr. Alicia Fornés is an Associate Professor at the Universitat Autònoma de Barcelona (UAB) and an affiliated researcher at the Computer Vision Center (CVC). Her research focuses on document image analysis, handwriting recognition and graphics recognition, with particular focus on historical documents and optical music recognition.

She has more than 100 scientific publications and has participated in numerous research and technology transfer projects. Her contributions have been recognized with several awards, including the AERFAI (IAPR) Thesis Award (2010), the IAPR/ICDAR Young Investigator Award (2017), the DonaTIC Award (2021), and the CIARP (IAPR) Aurora Pons-Porrata Award (2024).

During 2011-2020, she was in the Executive Committee of the IAPR TC-10 (Technical Committee 10 on Graphics Recognition), being the TC-10 Chair during 2019-2020. She has been the vice-president of the AERFAI (Spanish branch of the IAPR) during 2021-2024. She was also member of the commission "Mujer en Informàtica" (Women in Computer Science) from the SCIE (Spanish Scientific Society for Informatics) to analyze the current situation of women in computer engineering to establish synergies and to reduce the existing gap.