Abstract:
The problem of recognizing handwritten Gujarati characters has been tried by many researchers but still it requires enough work from vowel recognition up to its online application. The problem becomes even more complex if we use characters with vowels. Machine learning and Deep learning are extensively used to solve the image classification problem. It is observed by reviewing survey papers that Support Vector Machine, Bayes Probability Model, Deterministic Finite Automaton (DFA), Hidden Markov Model techniques are used as classifiers in this problem but machine learning and deep learning gives more promising result. It requires large data set to train and test the model. We collected hand-written Gujarati ‘Barakshari’ and text images from more than 1000 people having different ages. Deep learning requires a comparatively larger image set than machine learning. Both can have their pros and cons and so it is very much essential to optimize the data set if we are using the ‘hybrid mode’ to get benefits from both. Different augmentation techniques are also applied to the image set to raise the size, quality, and variety
