A New Approach of Deep Learning-Based Tamil Vowels Prediction Using Segmentation and U-Net Architecture

Deep Learning

Deep learning was developed mainly to process the images where the parameters will be huge especially when it comes to video segments. Furthermore, deep neural networks have paved a way through these by applying various algorithm models. The research related with any field like image processing, data analytics, Industry 4.0, Image segmentation, video segmentation, data visualization can go ahead with deep neural networks. This chapter will be mainly focusing on the processing of an image through deep neural networks. Like any machine learning model deep neural networks require training. It is essential to know about different kinds of neural networks and cost functions that can yield specific output models for text recognition.

In deep learning fundamentally the neural networks are classified into four main types as follows:

• ●

  Unsupervised pre-trained networks

• ●

  Convolutional neural networks

• ●

  Recurrent neural networks

• ●

  Recursive neural networks

In this chapter we will be only discussing supervised neural network models which is widely practiced in the present times.