Gabor filter based hand-drawn underline removal in printed documents

In this paper, a novel technique has been used for underline detection and removal in a Bengali and English document using Gabor filter and connected component analysis. In the underline detection module we first use Gabor filter in a specific direction to detect underline region and then use connected component analysis to detect particular underline. In second part i.e. underline removal module we use nearest neighbor approach. Our algorithm used for both touched and untouched underline. The experimental result of our proposed method demonstrates that it is able to improve Optical Character recognition (OCR) recognition accuracy.