Regresar

Improving word recognition with a time of flight 3D camera

Abstract:

- This study introduces a modification to a previous method for dewarping book spread images in the design of an automated book reader for persons with visual impairment and blindness. This design concept could also be applied to the challenging book digitization process. In particular, we will focus on contrasting the effects of using uniform and full height maps, obtained through a time of flight device, when performing the necessary image corrections. The experiments conducted to prove the merits of this approach were performed on a testing dataset consisting of 142 pages with their corresponding depth maps that were extracted using the time of flight 3D camera. These 3D maps of book spreads are made available to other researchers as an open source resource for developing other dewarping mechanisms and character recognition algorithms. The results were quantified and measured by introducing the corrected images to an Optical Character Recognition (OCR) engine. Lastly, the robustness of the approach utilizing these height maps (uniform and full height) is also put to test by introducing unforeseen rotation on the book spreads which could happen when the book is not place properly.