Digits Recognition of Marathi Handwritten Script using LSTM Neural Network

Penmanship acknowledgement is a famously troublesome issue in Machine Learning. In spite of many years of innovative work, current penmanship acknowledgement frameworks still show imperfect execution in reality applications. Ongoing examinations show extraordinary capability of Recurrent Neural Networks (RNN) with Long Short-Term Memory (LSTM) for transcribed content acknowledgement. In this paper we assess two methodologies dependent on RNN with LSTM for the acknowledgement of transcribed digits in Marathi Script. The methodology utilizes a Connectionist Temporal Classification yield layer. To test this methodology we have constructed a penmanship acknowledgment framework which takes a written by hand Marathi digit picture as an info and gives a decoded text (digit) as a yield. This methodology shows promising outcomes with 79.637% Digit-level precision on the test dataset for the strategy. The Connectionist Temporal Classification approach reliably beats the other Machine learning approaches as far as speculation and expectation of Complex digits.