Generative artificial intelligence technology provides new insights for art image inpainting and semantic reconstruction.To address the problem of semantically incorrect restoration content in existing research, this paper first optimises generative adversarial network by combining gated convolution with spectral normalisation.Based on this, an image inpainting and semantic reconstruction method is built by integrating text and art image features.The text disentanglement module of the suggested method can obtain key textual features that help restoration.A cross-modal attention module is designed to ensure restored results are as consistent as possible with text semantics.A dual-channel reconstruction module is also designed to enhance the network's ability to predict image structure and text semantics.Experimental results show that the frchet inception distance (FID) of the proposed method is 3.01, which can restore realistic art images satisfying textual semantics.
Huiling Huang (Thu,) studied this question.