Effective Optical Braille Recognition Based on Two-Stage Learning for Double-Sided Braille Image

Abstract

This paper proposes a novel two-stage learning framework TS-OBR for double-sided Braille images recognition. In the first stage, a Haar cascaded classifier with the sliding window strategy is adopted to quickly detect Braille recto dots with high confidence. Then a coarse-to-fine de-skewing method is proposed to correct original skewed Braille images, which maximizes the variance of horizontal and vertical projection at different angles. And an adaptive Braille cells grid construction method based on statistical analysis is proposed, which can dynamically generate the Braille cells grid for each Braille image. In the second stage, a decision-level SVM classifier with four classifiers recognition results is used to get recto dots detection results only on intersections of the Braille cells grid. Experimental results on the public double-sided Braille dataset and our Braille exam answer paper dataset show the proposed framework TS-OBR is effective, robust and fast for Braille dots detection and Braille characters recognition.