Scene Text Detection Using Pyramid-Based Text Proposal Network and Transformation Component Network
A. S. Venkata Praneel, Dr.T. Srinivasa Rao · Indian Journal of Computer Science and Engineering · 2023
Text identification and Text Recognition in Natural Scene Text importance is growing attention in recent research.Scene text contains a wealth of semantic information that can be used in a broad spectrum of vision-oriented applications.Text Recognition and Detection in the image of a natural scene play a vital role in accessing information and understanding the environment.Scene text orientations comprise horizontal, arbitrarily, curved, and vertically oriented scene texts.In this study, a Mask Scoring R-ConvNN based text detection strategy in this research can reliably detect curved text and multi-oriented from real input images.For learning good capability mask scores for the expected instance, a Mask Scoring R-ConvNN network frame is employed in this model.The mask scoring system corrects the misalignment between mask quality and score while also improving instance segmentation performance by emphasizing more accurate mask predictions.Using a Pyramid-based Text Proposal Network (PBTPN) and a Transformation Component Network (TCN) to improve the Feature extraction capabilities of Mask Scoring R-ConvNN for text identification and segmentation.Studies showed that Pyramid Networks are more successful at suppressing false alarms triggered by text-like backgrounds.By employing testing on a scale and with a single model, this strategy can obtain higher performance on Multi Oriented text and curved text benchmark datasets.