Enhancement and feature extraction for images of incised and ink texts

摘要

This paper describes the image enhancement techniques that we have used to help historians read three kinds of writing tablets: ink, wooden stilus and lead curse tablet. The techniques include: homomorphic filtering to correct uneven illumination, high-pass filtering to remove shading caused by surface undulations, and a newer technique to remove the often substantial complicating factor of wood grain. A phase-based feature detector has been developed to detect text stroke. A description of the detected strokes is generated for a text recognition system.