Handwritting Recognition Using Keras

Adisoka Bimantara, Alhamda (2018) Handwritting Recognition Using Keras. Technical Report. studentstaff5, Perpustakaan Institut Teknologi Telkom Purwokerto. (Unpublished)

	Text COVER.pdf Download (321kB)
Preview	Text ABSTRAK.pdf Download (7kB) | Preview
Preview	Text ABSTRACT.pdf Download (6kB) | Preview
Preview	Text BAB I.pdf Download (77kB) | Preview
	Text BAB II.pdf Restricted to Registered users only Download (93kB) | Request a copy
	Text BAB III.pdf Restricted to Registered users only Download (564kB) | Request a copy
	Text BAB IV.pdf Restricted to Registered users only Download (9kB) | Request a copy
Preview	Text DAFTAR PUSTAKA.pdf Download (69kB) | Preview
Preview	Text LAMPIRAN.pdf Download (353kB) | Preview

Abstract

On the practice of fieldwork in SOLUSI247 Yogyakarta project creation is handwritting recognition for recognize every handwritting in an image or photo. Handwritting recognition support REST API service which will be integrate with Chanthel program for support indexing and contenting. The programming language used is Python. Library or module used for prediction is Keras and Tensorflow. Next, OpenCV is for image processing. Then microframework Flask used to make web application that is frontend and REST API. The dataset used is around 280,372 data, which 70% as data training, and 30% others as data testing. The Model used is Convolutional Neural Network (CNN) with the accuracy result is around 79%. Handwritting recognition have several stages to predict such as starting from load model, load image for test, and then image procesing for improving image quality such as convert image to grayscale, sharpen image with threshold, increase the edge line of image with dilation, and finding contours in image to split sentences, words, and characters. And then do predict which will be result of the prediction shown in JSON format. With the accuracy of the model is around 79%, this handwritting recognition program is enough to predict every handwritting character in an image or photo.There is mistakes in prediction especially recognize the same character with other character. This happens because less variation in dataset make program can't predict character perfectly. Then microframework Flask is suitable for making REST API service. Besides the convienence and many features offered, Flask is so fast for making prediction and sending responses to client where the time needed is less than one second, just 859 miliseconds. And Image processing is very helpful in prediction process, especially on improving image quality. Keywords : Handwritting Recognition, Deep Learning, CNN, Keras, Flask, REST API, JSON

Actions (login required)

View Item