Method and device for outputting information

Abstract

The embodiment of the invention discloses a method and device for outputting information. A specific embodiment of the method comprises the steps of obtaining a style picture including to-be-replacedcharacters and a text picture including target characters; learning the text style attribute of the to-be-replaced text from the style picture; generating a foreground picture and a mask picture basedon the text style attribute and the text picture; erasing the characters to be replaced from the style picture to generate a background picture; and fusing the foreground picture, the mask picture and the background picture to generate a target picture and outputting the target picture. According to the embodiment, the advantages of target text style incremental learning are integrated, an adversarial generative network learning method is adopted, text geometrical characteristics and other attribute characteristics of any shape are learned in a multi-dimensional mode, and meanwhile the problem that text structural characteristics can be lost in the process of generating the target text through other methods is solved. The method is applied to image generation, the operation speed is high,and some real-time application scenes can be met.

Description

technical field [0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular to methods and devices for outputting information. Background technique [0002] Text in images and videos, or called scene text, contains rich semantic information and is very useful in many multimedia applications. The work related to text style transfer and background texture restoration has always been a hot topic of research by many scholars in the field of text editing. With the transformation of the deep learning era, text image editing has also changed from traditional methods to style transfer based on adversarial neural networks. [0003] In text editing, maintaining the original text style is a very challenging job. People try to transfer or copy the style information of their text in a given image or stylized text sample. Some current methods focus on character-level style transfer. For example, self-encoding network text style transfer an...

AI Technical Summary

Problems solved by technology

Therefore, it is difficult to directly apply text style transfer based on existing methods

At the same time, most of the current image generation methods use neural convolutional network image reconstruction methods, but the attribute information of the text is easily lost during the generation process, the structure of the generated results is irregular, and the font is inconsistent with the previous ones, etc.

Images