Research On Layout-to-Image Generation Based On Edge Enhancement

Deep learning technology has made significant progress in recent years and has been widely used in computer vision.As a branch of computer vision,conditional image generation task based on layouts has made remarkable achievements.However,in the images generated by the existing methods,pixel blending still exists at the boundary between various components,resulting in problems such as distortion and poor recognition of object-level generation.In response to the above problems,this dissertation mainly studies the layout-to-image generation methods focusing on edge enhancement,including edge information enhancement strategy,model design,edge dataset production,etc.According to different sources of edge information,this paper explores the importance of edge information to object quality improvement from two perspectives.In this paper,a layout-to-image generation algorithm based on edge extraction and feature information fusion and a Memory-based layout-to-image generation algorithm are proposed in turn.A large number of experiments have verified the advancement of the proposed algorithms.The algorithm proposed in this paper has important research value and significance for applying deep learning technology in layout-to-image generation.The innovation and contribution of this paper mainly include the following two aspects:1.A layout-to-image generation algorithm based on edge extraction and feature information fusion is proposed.The feature encodings of edge information are learned from the multi-level features output by the generator and iteratively optimized along the generator’s pipeline.Meanwhile,the discriminator is fed with frequency-sensitive image features,which significantly enhances the generation quality of the image’s high-frequency edge contours and low-frequency regions.Extensive experiments demonstrate that the method proposed in this paper outperforms the state-of-the-art methods,with more complete edge contours and less noise.2.A Memory-based layout-to-image generation algorithm is proposed.For the generator to explicitly focus on the edges of each component in the image,this paper constructs a Memory pool to store edge feature information.In image generation,the most similar edge features in the Memory pool are found according to the object feature vectors.These features are used as auxiliary information to participate in subsequent image generation task.Furthermore,the feature items in the Memory pool are obtained from the self-constructed edge dataset edge_COCO through the edge feature extraction module.A large number of experiments have proved the effectiveness of the proposed algorithm,and this method provides a new idea for the layout-to-image task.In generating images from layouts,this paper innovatively proposes to improve the quality of the object level by focusing on the edge information of each component in the image,and provides two solutions.This paper has carried out theoretical and practical explorations along this idea.The results show that edges are important and necessary for improving the quality of objects and images.It is hoped to bring some thoughts and inspiration to the subsequent related researchers.