Vernacular Language Handwriting Recognition Using Deep Learning Techniques

The work focuses on the detection and recognition of Handwritten Hindi Language using Convolutional Neural Networks and displaying it in printed form. The dataset used is derived from Kaggle, which comprises 96 thousand text images with a dimension of 32 × 32, out of which 80% of the data has been used in training and the rest 20% in testing the model. The dataset contains images of different types of handwritten Hindi alphabets and their respective values. After the various feature extractions, the real-time image processing is done to remove skewness and shadows and after character segmentation and removal of shirorekha, we finally get the final printed form of the Hindi text with 98.5% validation accuracy.