Text-independent chinese writer identification using hybrid SLT-LBP feature

Abstract

This study proposes a new hybrid method using texture features of input handwriting document image as global to overcome the limitation of data heterogeneity, which causing the ambiguity and leads to inconsistent results apart from problems of scale involve database size. The method first adopts Slantlet Transform (SLT) to bring out hidden texture details prior to feature extractions. Then, Local Binary Pattern (LBP) descriptor is applied on the SLT image to extract texture features. A new hybrid method Slantlet Transform based Local Binary Pattern (SLT-LBP), are experimented on an open and widely used HIT-MW Chinese database for performance evaluation. This study strengthens the idea that to unravel some of data heterogeneity and lead to improve identification performance, especially searching for relevant document from large complex repositories is an essential issue.