CHARACTER RECOGNITION

Abstract

Character recognition has numerous applications namely postal addresses, bank check amounts, documents, number plates, etc. Development of a recognition system is an emerging need for digitizing handwritten documents that use Tamil characters. Therefore, we created a prediction model to recognize Tamil characters, the model works based on the Deep convolutional neural network (CNN). In this project, model is sequential. It is like a derivation of LeNet-4 architecture, including a dropout layer.A dataset for few Tamil characters is created, along with the recognition model. It helps people in identifying the exact script written in document, where some may find it difficult to recognize the characters written in paper due to colloquial writing and different handwriting styles. CNN have given excellent results to old fashioned shallow networks in acknowledgement tasks. To avoid overfitting in the recognition model, accuracy is increased by using dropout layer and dataset increment method. With the help of these methods in the CNN model, test accuracy rate was increased. The modifiedCNN architecture achieved a highest test accuracy of 97.75%.