Anisotropic diffusion with edge projection

Abstract

In image processing, anisotropic diffusion provides a forward method to remove noise while preserving edges accurate and sharp. However, due to the inappropriate edge estimation by gradient, some isolated noise points still exist and edge location is inaccurate. In this representation, isolated noise points and edges are distinguished by the significant difference of their “lengths”, which are computed by orthogonally projecting their pixels to the corresponding normalized gradient directions and recording the number of the same projections. Combining gradient and “length” to estimate edges, isolated noise points are further suppressed while edges are re-located and enhanced.