An Improved Offline Text-independent Chinese Writer Identification Scheme based on Two-tier Image Retrieval Mechanism

Abstrak

Research in writer identification has received significant interest in recent years due to its forensic applicability.  Undoubtedly, many achievements have been carried out on the traditional method which is without retrieval and only focused on inconsistent and lead ambiguous identification performance.  A major problem with this kind of traditional method is searching and retrieval of a document from large image repositories is currently a big issue.   In this paper, the focus aim is to determine the effectiveness and reliability of integrating retrieval mechanisms compared to the best and up-to-date techniques for writer identification without retrieval mechanism in offline text-independent Chinese writer identification.  Experiments were conducted on an open HIT-MW database which is widely used for performance evaluation and employed the same standard dataset for benchmarking. The proposed method incorporates a combination of selected features—Statistical Local Ternary Local Binary Pattern (SLT-LBP), Histogram of Contour (HC), and Gray Level Difference Method (GLDM)—integrated with a Euclidean distance-based classification framework. Experimental evaluations conducted on the publicly available HIT-MW dataset demonstrate that the proposed approach achieves an identification accuracy of 96.68%.  These results indicate the potential of the proposed method to perform competitively with existing state-of-the-art techniques, while also offering improvements in scalability and interpretability for writer identification tasks.  Integration method with two-tier image retrieval for reducing search space in interpretability of results by forensic experts when large databases are involved and improving identification rates, yet remarkable accuracy.  This area, however, still has a large room for research which can be taken by upcoming researchers.