Interactive Enhancement of Handwritten Text through Multi-resolution Gaussian

Abstract

Handwritten text found in historical documents is often difficult to read due to issues such as contrast, noise and degradation. There has been much work on how to process such documents including improvements on binarization of these images. Despite the different advances in this area, improving the quality and readability of these documents is still an open research area. In this paper a novel approach is proposed to improve the text of historical documents through interactive stroke enhancement. This approach utilizes user interaction to indicate parts in the image where stroke enhancement is needed. The algorithm uses a difference of multi-resolution Gaussians to detect text at different scales and to modulate the amount of enhancement needed. This approach could be used for manually restoring text images or for improving readability of the text. Results are given in this paper that show the effectiveness of the proposed method.