Degraded document image preprocessing using local adaptive sharpening and illumination compensation

Preprocessing for degraded document images can improve their binarization result. Sharpening and illumination compensation are effective methods in preprocessing. We find the degree of sharpening has different effects on the different stroke width. As the degree of sharpening increases, the thin strokes retained more information in the binarization results, while the thick strokes gradually appear to be broken. Aiming at this problem, we propose a local adaptive sharpening method. The stroke width estimation algorithm is utilized to estimate the stroke width in the local region. Local adaptive sharpening is performed to solve the problem of thick stroke fracture and retain more information of thin strokes. In addition, comparing with the weakly illuminated document images, the sharpening effects on strongly illuminated document images are more prominent, and the binarization result is better. Therefore, appropriate illumination compensation is used for weakly illuminated document images. We further propose a preprocessing method for degraded document image using local adaptive sharpening and illumination compensation. The experimental results show that our proposed method restores more detail information and keeps the thick stroke information in binarization result. Our method outperforms U-Net without preprocessing by 0.36% FM scores on DIBCO2016, 1.09% FM scores on DIBCO2017 and 1.42% FM scores on DIBCO2018. U-Net, Sauvola and OTSU combined with our LASIC outperform themselves by 1.42%, 0.29% and 5.41% FM scores on DIBCO2018. And our LASIC method outperforms other preprocessing methods by 0.1% to 1.05% FM scores on DIBCO2016-DIBCO2018.