Text-to-Image algorithm Based on Fusion Mechanism

What is it about?

This study proposes a multi-step text-generated image algorithm that utilizes feature fusion to address the issues of image blurring and missing image details in text-generated images. To improve the ability of single-channel text features to guide multi-channel image features, the study implements a feature fusion module that migrates text features to image features while refining the details of generated images. The fusion module is alternately executed with the upsampling operation in the generator to increase the frequency of text feature usage. Placing the generator and discriminator in three pairs achieves the goal of generating clear large images from fuzzy images. Experimental data on the CUB dataset show improvements in the Inception Score and Frechet Inception Distance score, and comparative analysis of the generated images indicate rich detail texture and sharpness. This research can provide insights for further development of text-generated image technology and can be applied to various fields requiring high-quality image generation such as visual effects production and natural language processing.

Featured Image