Road Scene Text Detection and Recognition Using Machine Learning
Syed Hassaan Ali Shah, Jamal Hussain Shah · 2023
For high-level semantic information extraction, text is the most powerful source. Understanding natural scene text is a hot topic in the area of computer vision. It means that natural scenes encapsulated a variety of specific information in the form of text, which may be applied in a variety of applications of real world. This study will look at all areas of scene text comprehension while adding fresh machine learning algorithms in a semi-pipelined or fully pipelined methodology. The primary target of this study is to produced and implement integrated algorithms that automatically perform image processing and computer vision techniques in order to understand, correct, and overcome all text-related challenges under a single umbrella, resulting in an all-in-one end-product. In the first phase of this research work an object detection model i.e. YOLOv5 is applied on ASAYAR dataset to localize the road scene text images in the form of bounding boxes. This model has worked more effectively and produced accuracy level up to 99% on text images. In the second phase preprocessing techniques are applied and quality of image dataset is enhanced by using K-Means color segmentation. The enhanced images are then passed through Maximally Stable Extremal Region (MSER), which is a feature region detector for text based images. After detection of text regions, Optical Character Recognition (OCR) is applied for the final text recognition.