Text Spotting Software
Text spotting, namely the ability to read the text in natural scenes, is a highly desirable feature in anthropocentric applications of computer vision. The software utilises a Fully-Convolutional Regression Network (FCRN) for high-performance text recognition in natural scenes which detects text up to 45× faster than the current state-of-the-art text detectors and with better accuracy.
The FCRN is trained without any manual annotation using a new dataset of synthetic text in the wild. The latter is obtained by automatically adding text to natural scenes in a manner compatible with the scene geometry.
Text detection
This new method and software make two new contributions to the current state of the art. First, it utilises a new method for generating synthetic images of text that naturally blends text in existing natural scenes, using off-the-shelf deep learning and segmentation techniques to align text to the geometry of a background image and respect scene boundaries.
Second, it implements a text detection deep architecture which is both accurate and efficient. The combination of these two allows the system to exceed state-of-the-art-performance.
about this technology
