NFP-AST:自然特征保留的任意风格迁移模型

NFP-AST:arbitrary style transfer model for natural feature preservation

亲和特征提取和自然融合是实现风格迁移的关键。为此,提出一个新的自然特征保留的任意风格迁移模型NFP-AST。通过可逆残差网络在前向和后向推理中对特征二分处理,保证了提取特征亲和性,减少因提取过程造成的图像重建误差。在自适应空间重构模块ASRM中,先通过全局统计信息匹配内容风格特征,接着在融合特征中插值自适应权重捕获细节无偏融合内容风格特征,使风格过渡自然。定性和定量实验研究结果表明,NFP-AST产生的风格化图像与先进方法相比都取得了较好得分,艺术表达更具有视觉冲击力。

The extraction of affinity features and natural fusion are crucial for achieving style transfer.To address this problem,this paper proposed a novel arbitrary style transfer model,called NFP-AST,emphasizing the preservation of natural features.Through a reversible residual network,it performed binary processing of features in both forward and backward infe-rences,ensuring the extraction of affinity features and reducing image reconstruction errors.In the adaptive space reconstruction module(ASRM),it firstly used global statistical information to match content and style features,followed by interpolating adaptive weights in the fusion features to capture details for unbiased merging of content and style features,resulting in a natural style transition.Qualitative and quantitative experimental results indicate that NFP-AST produces stylized images with better scores compared to state-of-the-art methods,demonstrating enhanced visual impact in artistic expression.