Tai Lue handwriting recognition via multi-channel feature fusion and deep Gaussian process

Although deep learning has achieved significant progress in handwriting recognition, systematic studies on handwritten New Tai Lue character recognition remain limited. This task is inherently challenging due to casual writing styles, structural irregularity, and high inter-class visual similarity, where discriminative cues often lie in subtle stroke-level variations. To address these challenges, we propose a recognition framework based on multi-channel feature fusion and a mixed-kernel deep Gaussian process, MK_DGP. A handwritten New Tai Lue dataset is first constructed to support systematic evaluation. The fusion mechanism integrates local and global structural features, while MK_DGP models complex decision boundaries to enhance robustness. Experimental results demonstrate that the proposed method achieves 99.18% accuracy and outperforms existing approaches.