Handwritten Devanagari Script Segmentation: A non-linear Fuzzy Approach
This work addresses segmentation challenges in handwritten Devanagari script, which is important for optical character recognition in languages like Hindi, but it is incremental with a 1.8% improvement over prior methods.
The paper tackled the problem of segmenting handwritten Devanagari script by developing a non-linear fuzzy approach for identifying Matra pixels and segmentation points, achieving a segmentation accuracy of 94.8%.
The paper concentrates on improvement of segmentation accuracy by addressing some of the key challenges of handwritten Devanagari word image segmentation technique. In the present work, we have developed a new feature based approach for identification of Matra pixels from a word image, design of a non-linear fuzzy membership functions for headline estimation and finally design of a non-linear fuzzy functions for identifying segmentation points on the Matra. The segmentation accuracy achieved by the current technique is 94.8%. This shows an improvement of performance by 1.8% over the previous technique [1] on a 300-word dataset, used for the current experiment.