A Novel System Model to Recognition of Merged-Characters Under Linear, Nonlinear and Overlapped

This article introduces a traditional model for recognition of the merged-characters. It consists in modeling each sequence with the segmentation which is to cut the isolate items and the recognition which is to make the decision for accepting. In this paper we propose two algorithms based on this system model with one Kinematical Drop-Fall Algorithm for segmentation which could resolve different kinds of merged-relationship as Linear, Nonlinear and Overlapped summarized by Song et al. and a revised Complex Moment Invariants for recognition of single object under the translation, scaling and rotation (TRS). These frame of model and two aspects of improvements in the model could provide two major benefits: Firstly, through the improvement in two aspects, the final effect in correctness and intactness is guaranteed. Moreover, the segmentation and recognition containing in the common set offset and help each other so that would be more effective to obtain the better performance. The experimental results show that our proposed revised algorithms for segmentation, recognition and whole system model are all receiving the better performances compared to former works from the Civil Aerial Meteorological Maps and UPS normalized characters dataset.