Efficient multiscale and multifont optical character recognition system based on robust feature description

Optical Character Recognition (OCR) is the process of translating images of text into a comprehensible machine format. Generally, an OCR system is composed of binarization, segmentation and recognition stages. Given an extracted binary character, the recognition stage ensures its description and decides its corresponding ASCII code. In this paper, we propose a new OCR system that aims to high speed, Multiscale and Multifont character recognition. Our proposal is based essentially on robust description using a new Unified Character Descriptor (UCD). In addition, a character type-face and font-size recognition is performed to choose the adequate template for faster matching process. Obtained OCR Accuracy of our proposed System is 1.5x higher then that reached by Tesseract on the LRDE dataset.