高级检索

基于语义分离和特征融合的人脸编辑技术

An Independent Semantic and Fused Latent Model for Local Face Editing

  • 摘要: 人脸图像编辑模型中的语义属性之间有较强的关联性, 编辑其中一种语义可能导致其它语义属性以及非编辑区域的内容改动. 为了提升用户的编辑体验, 实现对人脸图像细节更为精确的编辑, 本文提出了一种在图像域上语义分离和特征融合(Independent Semantic and Fused Latent, ISFL) 的人脸图像编辑模型. 首先使用图像掩码将人脸图像的各个语义分离, 并将人脸语义组织成一个层次化的树状结构, 然后通过ISFL实现对图像语义的局部分离和全局融合, 用户可通过掩码单独编辑图像中不同语义的结构和外观; 最后使用了基于编码器和基于优化的两种方式对生成图像进行了细节部分的优化. 实验表明本文提出的图像编辑方法可以得到更加真实, 细节更加丰富的图像.

     

    Abstract: There is a strong correlation between semantic attributes in face editing models; editing one attribute may result in changes to other semantic attributes and non-edited areas. To enhance user editing experience and achieve more precise editing of facial details, this paper proposes a face editing model based on semantic separation and feature fusion in the image domain, termed Independent Semantic and Fused Latent (ISFL). Firstly, facial image semantics are separated using image masks and organized into a hierarchical tree structure. Then, ISFL is used to achieve local separation and global fusion of image semantics. Users can independently edit the structure and appearance of different semantics in the image using masks. Finally, two methods, encoder-based and optimization-based, are employed to optimize the details of the generated image. Experimental results demonstrate that the proposed image editing method can produce more realistic and detail-rich images.

     

/

返回文章
返回