摘要:
当今深度超分网络参数量非常庞大, 且推理速度缓慢. 而轻量化网络在复杂环境条件下存在无法表达图像深层特征的问题. 针对以上问题, 提出基于特征增强的轻量化可逆超分辨率网络. 首先, 提出边缘特征残差, 配合所提的边缘相似损失指导模型重建, 增强重建图像对纹理轮廓的表达能力; 然后, 补充新的小波特征核, 使小波变换支持任意缩放因子的重建任务; 最后, 引入全局特征提取模块, 在特征图中嵌入自注意力机制, 提取全局特征. 在缩放因子为4时的基准测试集Set5上的实验结果表明, 所提出的网络相比SwinIR-light表现更优, PSNR提升0.41, 参数量减少244K, 推理时间降低49.05%.
Abstract:
Deep super-resolution reconstruction networks have a large number of parameters and slow inference speed. Lightweight networks unable to express deep features of images under complex environmental conditions. Aim-ing at those issues, a lightweight reversible super-resolution network based on feature enhancement is proposed. Firstly, edge feature residuals are proposed, combined with the proposed edge similarity loss to guide model re-construction and enhance the expression of texture contours; Then, add a new wavelet feature kernel to support the reconstruction task with any scaling factors; Finally, a global feature extraction module is introduced to em-bed a self attention mechanism in the feature map to extract global features. The proposed network showed bet-ter performance than SwinIR-light on the benchmark test Set5 with a scaling factor of 4. It improved the PSNR by 0.41, reduced the number of parameters by 244K, and lowered the inference time by 49.05%.
下载:
