Publications by authors named "Marcal Rusinol"

Article Synopsis
  • Handwritten Text Recognition shows strong results in benchmarks but requires large labeled training datasets due to the variety in handwriting styles.
  • To reduce the need for extensive manual labeling, synthetic data from TrueType fonts is often used, although it introduces a style bias that affects recognition accuracy.
  • To overcome this, a new generative method is introduced that creates diverse handwritten text images based on visual appearance and text content, which has been shown to significantly enhance recognition performance through experimental validation.
View Article and Find Full Text PDF