Offline Handwritten Text Recognition (HRT) is a dynamic area of research focused on transcribing handwritten text from images. While humans can often decipher such text with ease, automating this task poses several challenges. In our study, we employ two distinct datasets: the Dead Sea Scrolls (DSS) and IAM [1] collections. These datasets vary in data type and quality, necessitating separate approaches.
The DSS dataset comprises aged Hebrew text that has deteriorated over time and is not segmented. However, only individual letters are provided as labeled data, requiring first the segmentation of the original text images into single characters before the recognition process can take place. On the other hand, the IAM dataset features images of handwritten lines of English text accompanied by the respective transcriptions as labels. This allows for the implementation of an end-to-end system. Our work presents methodologies for addressing both tasks, catering to the unique characteristics of each dataset.
The current group projects were implemented in the context of the course "Handwriting Recognition" taught by Professors Lambert Schomaker and Maruf A. Dhali at University of Groningen. For a comprehensive overview of the methodologies and final results, please refer to the Report. Additionally, a separate discussion section, offering a personal perspective, is available here.
1st Assignment: Character Segmentation and Recognition on the DSS Collection
- Project Description (Task 1 & 2)
- Implementation
- Report (Section 3)
2nd Assignment: Line Recognition on the IAM Dataset
- Project Description (Task 3)
- Implementation
- Report (Section 4)
[1] U. Marti & H. Bunke (2002). The IAM-database: An English sentence database for offline handwriting recognition. International Journal on Document Analysis and Recognition. vol. 5 (pp. 39-46). DOI: 10.1007/s100320200071.
- Chryssa Nampouri
- Shray Juneja
- Luca Mueller
- Wopke de Vries
