Noise removal from binary patterns by using line adjacency graphs

This paper proposes heuristic procedures to remove noise from character patterns. Most noise removal procedures described in the literature, succeed in removing only small isolated pieces of noise, called preliminary noise. These algorithms determine whether a given pixel is salt or pepper noise based on the local distribution of the pixels. Our procedures can remove larger chunks of noise (called postliminary noise) such as holes and specks. For our procedures to be effective, we employed both black line adjacency graphs (LAGs) and white LAGs, where white LAGs are our extension of the well-known black LAGs. By doing so, we have experimentally observed that, the noise-free image of the pattern is of a far better quality