Deep Convolutional Sequence Approach Towards Real-Time Intelligent Optical Scanning

Abstract

Visual text recognition is the most dynamic computer vision application due to its rising demand in several applications like crime scene detection, assisting blind people, digitizing, book scanning, etc. However, numerous research works were executed on static visuals having organized text and on captured video frames in the past. The key objective of this study is to develop the real-time intelligent optical scanner that will extract every sequence of text from high-speed video, noisy visual input, and offline handwritten script. The scientific work has been carried out with the combination of multiple deep learning approaches, namely EAST, CNN, and Bi-LSTM with CTC. The system is trained and tested on four public datasets (i.e., ICDAR 2015, SVT, Synth-Text, IAM-3.0) and measured on the basis of recall, precision, and f-measure. Based on the challenges, performance has been examined under three different categories, and the outcomes are optimistic and encouraging for future advancement.