Crimson Publishers Publish With Us Reprints e-Books Video articles


COJ Robotics & Artificial Intelligence

Text Detection using Object Recognition Techniques

Submission: February 19, 2019;Published: March 12, 2019


In this paper we propose an approach in text detection using object detection technique. Our approach is to deal with letters as objects. We use an object detection method, Retina Net deep leaning, to detect letters and recognize the text in natural scene images. The goal is to achieve high accuracy in text spotting, especially for curved text where the state-of-art methods fail. The Retina Net model architecture was used and modified in different ways to find the best performing model. Retina Net is implemented using Keras with Tensor flow backend. We prepare a model for training using the total-text dataset. The system manages to detect the letters as objects in the images. Then we perform letter candidate grouping to detect text based on distances between neighboring letters. The results exceeded our expectations. The training data was barely sufficient for some letters due to the differences in the frequency of their appearance in the dataset. Moreover, some specific letters, in specific orientations, were confusingly identified as others with a similar pattern. Nevertheless, we obtained a good performance on the test data for some classes, with a mAP of 40%. In order to further develop this method and to improve performance, more training data is needed, containing letters in different fonts. We also consider adding a dictionary to help correct or complete missing letters

Keywords: Text detection; Object detection; Deep learning; Retina net; Total text

Get access to the full text of this article

About Crimson

We at Crimson Publishing are a group of people with a combined passion for science and research, who wants to bring to the world a unified platform where all scientific know-how is available read more...

Leave a comment

Contact Info

  • Crimson Publishers, LLC
  • 555 Madison Avenue, 5th floor
  •     New York, NY 10022, USA
  • +1 (929) 600-8049
  • +1 (929) 447-1137