Offline Handwritten Script Recognition Based on Texture Descriptors

The recognition of different script systems is a challenging task in the context of document image analysis. In this paper, we develop and evaluate an approach to the script recognition problem using writing as texture information to represent a manuscript class. Distinct descriptors are tested and analyzed. As main contributions, we proposed the use of a texture compression scheme, a partitioning of the document into blocks, as well as their further combination. According to our experiments, high accuracy rates are achieved for the script recognition problem when compared to other approaches available in the literature.