MT3S: Mobile Turkish Scene Text-to-Speech System for the Visually Impaired

TL;DR

This paper presents MT3S, a mobile Turkish scene text-to-speech system for the visually impaired that combines fast multi-scale text detection with OCR to enable real-time reading on mobile devices.

Contribution

The paper introduces a novel mobile system for Turkish scene text reading that is faster and maintains high OCR accuracy compared to existing systems.

Findings

System operates in real-time on mobile devices

Achieves OCR accuracy comparable to state-of-the-art

Demonstrates significant speed improvements

Abstract

Reading text is one of the essential needs of the visually impaired people. We developed a mobile system that can read Turkish scene and book text, using a fast gradient-based multi-scale text detection algorithm for real-time operation and Tesseract OCR engine for character recognition. We evaluated the OCR accuracy and running time of our system on a new, publicly available mobile Turkish scene text dataset we constructed and also compared with state-of-the-art systems. Our system proved to be much faster, able to run on a mobile device, with OCR accuracy comparable to the state-of-the-art.