Hierarchical Retrieval of Ancient Chinese Character Images Based on Region Saliency and Skeleton Matching

Ancient Chinese characters exhibit characteristics of complex structure and diverse styles. To address the limitations of traditional Chinese character image retrieval techniques when applied to ancient Chinese characters, this paper proposes a hierarchical retrieval method for ancient Chinese characters images based on region saliency and skeleton matching (RSSM). The proposed method utilizes saliency joint weighting algorithm to effectively integrate the channel and spatial dimension information of deep convolutional features, enhancing the representation of key features. It focuses on capturing the detailed features of Chinese character contours and spatial structure, enabling coarse-grained retrieval of ancient Chinese characters. Furthermore, to further enhance retrieval accuracy, an improved shape descriptor, skeleton context, is introduced for fine-grained matching. The retrieval results are organized in ascending order of matching cost. The study constructs an ancient Chinese character image dataset named GJHZ. The Precision and mAP of RSSM achieved \(90.71\%\) and \(90.59\%\) , respectively. Experimental results demonstrate the superior performance of our method for ancient Chinese character image retrieval.