Image stylization aims to alter the visual features and textures of an image while preserving its core objects, structures, and semantics. Current methods have already focused on achieving image stylization using region-aware approaches. However, the style transfer results rely heavily on segmentation performance or semantic information. It is an important challenge to achieve finer regional control and more natural style fusion. To address this issue, this paper proposes an image Style Transfer method based on Similarity metric-guided Region Aware Transformer (S \(^2\) TRAT). The method optimizes the image stylization regionally by obtaining the semantic correlation between the content and style images through similarity metrics rather than semantic segmentation. Radical Basis Function (RBF)-guided cross-attention module is proposed to model the transformer guided by similarity metrics. Meanwhile, multi-layer attention modules are integrated into the short-range branch to enhance the ability of the S \(^2\) TRAT to capture local details and edge textures. Experimental results on image stylization on MS-COCO [1] and the style dataset WikiArt [2] demonstrate that our proposed method outperforms the state-of-the-arts methods in terms of subjective and objective results.