<![CDATA[Braille is a tactile writing system used to assist individuals with visual impairments. While Braille paper is a common medium, it is highly vulnerable to physical degradation, making automated optical recognition challenging. Despite effectiveness of local thresholding for degraded documents, grid search parameter analysis and configurations for Braille dot detection remain largely unexplored, leading to suboptimal detection performance. This study aims to systematically grid search Sauvola parameter analysis binarization parameters for Braille dot detection prior to Circle Hough Transform, encompassing a comparative evaluation against Niblack and Otsu baseline methods. To eliminate evaluation bias, dataset of 38 manually annotated 640x640 pixel images was strictly partitioned into a 10-image tuning set and a 28-image test set. Parameter grid search identified an optimal spatial boundary at window size=13. Mathematically aligning with maximum topographical footprint of a Braille cell and sensitivity parameter of k=0.050. In isolated test set, Sauvola achieved a superior Mean F1-Score of 0.7916, significantly outperforming Niblack of 0.7088, global Otsu thresholding of 0.4513, and CHT-only baseline of 0.7392. Our results suggest that normalization factor may contribute to observed performance differences. However, its individual effect was not isolated in present experiments.]]>
Copyrights © 2026
