A Joint Semantic-Based Approach to Text-Guided Domain Migration Image Translation

This paper presents a joint semantic-based image translation method for text-guided domain migration. Initially, a visual coder and visual text encoder generate a text description of the original image. This description, along with the target text, is fed into a text encoder to extract corresponding text embedding. Subsequently, a graphic semantic fusion network layer, based on the attention mechanism, generates fusion embedding representing a combination of image content structure and target text information. These embedding serve as a diffusion model for textual conditional guidance in image generation. For background information retention, a semantic segmentation algorithm produces a mask map of the background region, which is injected as a structural condition through a multilevel encoder to guide the translation process. The method's effectiveness is demonstrated through comparison experiments, ablation experiments, and image analysis of translation results.