Radical-based approaches for Chinese character recognition (CCR) have been extensively investigated and demonstrated substantial improvements. Chinese characters are generally decomposed into structural components and radicals, with recognition being accomplished through the accurate identification of these elements. However, the inherent complexity and diversity of Chinese radicals, combined with variations in writing styles, introduce ambiguities, potentially leading to suboptimal model performance. Such challenges are particularly evident when distinguishing radicals with visually similar characteristics. To address these limitations, this paper proposes a novel Adaptive Radical Similarity Measurement (ARSM) module, which enhances the robustness and generalization of CCR models by dynamically measuring the similarity between radical pairs and integrating this information with radical features. ARSM enables more robust feature integration and facilitates the development of a specialized loss function tailored for optimizing model parameters, thereby improving recognition accuracy. Comprehensive experiments conducted on multiple benchmark datasets demonstrate the effectiveness and robustness of the proposed method.