A new method for segmentation of pre-detected Devanagari words from the scene images: Pihu method

To date, only one method has reported on the segmentation of pre-detected Devanagari words from natural scenic images. The present paper discusses the limitations of the existing method and proposes a new robust method called the “Pihu method” (Pihu is the lovely daughter of the first author). The efficiency of the proposed method over the existing approach is demonstrated by considering a challenging dataset of natural scenic images of Devanagari script. The study shows that the character segmentation accuracy of the Pihu method was 92.11% and that of the existing method was 55.77%. The results show that it takes the Pihu method 2.70 s to segment an image, whereas the existing method takes 4.76 s.