Letter Recognition from Noisy Images Using Deep Learning

Abstract

Machine learning and deep learning are being applied in multiple fields. Letter recognition presents a very active and challenging research area. Current research demonstrates a marked increase in focus on deep learning methodologies applied to letter recognition. Letter recognition is a problem that's been studied extensively in various languages. Letter recognition refers to the skill of recognising and distinguishing letters based on their visual features. It includes examining response patterns and identifying particular elements that make up various letters. This perceptual process is essential for identifying distinct letters and is frequently examined through quick visual presentations and visual search activities. Furthermore, understanding letters in context explores how letters are perceived within words, thereby improving our understanding of reading processes. During the recognition process, pre-processing methods enhance image quality by minimising noise and adjusting orientation, whereas convolutional neural networks are responsible for extracting features of the letters. The current letter recognition system encounters numerous difficulties in extracting text from noisy and distorted images or complex layouts, with extraction primarily restricted to numerical characters and the English alphabet. Many studies have employed deep learning models to enhance accuracy. The proposed method reaches an impressive 96.2% accuracy in identifying letters from input images.