基于推理注意力机制的二阶段网络图像修复

Two-stage Network Image Inpainting Based onReasoning Attention Mechanism

现有的图像修复方法在处理大面积缺失或高度纹理化的图像时,通常会产生扭曲的结构或与周围区域不一致的模糊纹理,无法重建合理的图像结构。为此,提出了一种基于推理注意力机制的二阶段网络图像修复方法。首先通过边缘生成网络生成合理的幻觉边缘信息,然后在图像补全网络完成图像的重建工作。为了进一步生成视觉效果更逼真的图像,提高图像修复的精确度,在图像补全网络采用推理注意力机制,有效控制了生成特征的不一致性,从而生成更有效的信息。所提方法在多个数据集上进行了实验验证,结果表明该图像修复方法的结构相似性指数达到了88.9%,峰值信噪比达到了25.56 dB,与现有的图像修复方法相比,该方法具有更高的图像修复精确度,生成的图像更逼真。

When the existing image inpainting methods process large-area missing or highly textured images,they usually produce distorted structures or fuzzy textures that are inconsistent with surrounding areas,and cannot reconstruct a reasonable image structure.Therefore,this paper proposes a two-stage network image inpainting method based on reasoning attention mechanism.Firstly,the edge generation network generates the reasonable illusion edge information,and then the image completion network finishes the image reconstruction.In order to get more realistic images with visual effects and improve the accuracy of image inpainting,the reasoning attention mechanism is adopted in the image completion network to effectively control the inconsistency of generated features.The proposed method is validated by experiments on multiple datasets,and the results show that the structural similarity(SSIM)index and the peak signal-to-noise ratio(PSNR)can reach 88.9%and 25.56 dB,respectively.Compared with existing image inpainting methods,this method has higher inpainting accuracy and more realistic images.