Cross Domain Evaluation of Text Detection Models

• Published in: Artificial Neural Networks and Machine Learning - ICANN 2022 Vol. 13531; pp. 50 - 61
• Main Authors: Ali-Gombe, Adamu, Elyan, Eyad, Moreno-García, Carlos, Jayne, Chrisina
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer 2022; Springer Nature Switzerland
• Series: Lecture Notes in Computer Science
• Subjects: Character Aware Region Awareness for Text Detection; Efficient and Accurate Scene Text Detector; Ensembles; Tesseract; Text detection
• ISBN: 3031159330; 9783031159336
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-031-15934-3_5

Text detection is a very common task across a wide range of domains, such as document image analysis, remote identity verification, amongst others. It is also considered an integral component of any text recognition system, where the performance of recognition tasks largely depends on the accuracy of the detection of text components. Various text detection models have been developed in the past decade. However, localizing text characters is still considered as one of the most challenging computer vision tasks within the text recognition task. Typical challenges include illumination, font types and sizes, languages, and many others. Furthermore, detection models are often evaluated using specific datasets without much work on cross-datasets and domain evaluation. In this paper, we present an experimental framework to evaluate the generalization capability of state-of-the-art text detection models across different application domains. Extensive experiments were carried using different established methods: EAST, CRAFT, Tessaract and Ensembles applied to various publicly available datasets. The generalisation performance of the models was evaluated and compared using precision, recall and F1-score. This paper opens a future direction in investigating ensemble models for text detection to improve generalisation.