In order to address the problem of low accuracy in single-character detection caused by complex backgrounds, densely distributed small characters, and irregular shapes in oracle bone rubbing images, a method integrating image preprocessing and an improved YOLOv11s detection model is used to investigate single-character detection and candidate-region extraction in oracle bone rubbings. The results show that the proposed method achieves an mAP50 of 0.8930 and an mAP50-95 of 0.5700 on the open-source dataset. The Precision, Recall, and F1 scores are 0.8555, 0.8580, and 0.8567, respectively. Compared with the baseline model, the improved model performs better in dense small-object detection and localization of characters with complex deformations. It is concluded that the proposed method provides effective technical support for automatic single-character detection and candidate-region extraction in oracle bone rubbing images.