An Intelligent Technique for the Classification of Consonants and Vowels in Devnagari Script

Abstract

Convolution neural networks (CNN) and probabilistic neural networks (PNN) have taken considerable steps to recognize handwritten characters of consonants and vowels in the Devanagari script. The main objective of this work is to use the benefits of CNNS and PNNS to solve difficulties caused by the complex structure of the figures of Devanagari. CNN, known for its ability to extract spatial and hierarchical information, is particularly good for observing complex patterns and small variations between characters, which is ideal for this task. On the contrary, PNN uses probabilistic methodology depending on statistical metrics to effectively categorize characters, guaranteeing rapid convergence and high accuracy. The study examines the relative efficiency and usability of these approaches to recognize Devanagari scripts by merging them. A large collection of handwritten figures of Devanagari, including vowels and consonants, is used for the experiment. To improve the quality and variability of data, the data set has undergone extensive pre-process, which included enlargement and normalization. With a small medium square error (MSE) and good correlation coefficients, the CNN model showed exceptional capabilities of elements extraction, reached more than 98% of training accuracy and over 96% accuracy on validation and test data sets. Similarly, because the PNN model is non -neural, it did exceptionally well and gained the same accuracy with less training time. These findings show how well the two models of the complexity of the Devanagari script are suitable.