To address the recognition challenges in mixed-script Korean classical texts containing Chinese and Korean characters, this study proposes a unified character recognition method for ancient Korean documents named CKR-UniIDS based on unified encoding of Chinese and Korean characters. Firstly, considering the multi-script characteristics of classical Korean texts, we implement unified encoding for Chinese radicals, Korean letters, and 12 ideographic description characters based on structural similarities between Chinese and Korean characters. Secondly, to resolve the information loss caused by low-resolution text images in classical Korean documents, we employ multi-layer Spatial-Depth convolutional blocks (SPDConv) to mitigate severe information degradation during downsampling. To accommodate the rich stroke diversity of Chinese and Korean characters, we introduce linear deformable convolution (LDConv), whose adaptable kernel size and shape enable extraction of fine-grained stroke features within smaller receptive fields, thereby improving robustness to stroke variability. Finally, to address the limited samples and class imbalance in Korean ancient datasets, we implement a transfer learning strategy combining pre-training with Unicode font images and fine-tuning on ancient document datasets. Experimental results demonstrate that our method significantly outperforms existing approaches in zero-shot recognition of Korean classical texts, achieving a maximum accuracy of 44.94% on Korean classical datasets,28.98% higher than Liu’s method. After fine-tuning, our model reaches 73.19% accuracy with notable improvements under limited training data conditions. Additionally, the CKR-UniIDS model comprehensively surpasses radical/stroke-based methods and their combinations in recognizing artistic Chinese typefaces under zero-shot scenarios.