End-to-end character image generation method guaranteeing consistent style

Abstract

The invention discloses an end-to-end character image generation method guaranteeing consistent style. The method comprises the following steps: step 1, erasing characters in a source style image, supplementing background textures, and obtaining a background image without characters; step 2, rendering a target text into an image, embedding the image into the background image without characters output by a background reconstruction module, and migrating text styles of the source image into the target text under the same background. By adopting the end-to-end method, the work such as text style migration and character erasing is integrated into one network, so that the potential losses of effectiveness and performance caused by intermediate steps are reduced; by using a GAN algorithm, a result that is more real, more consistent in style, and more coherent in semantics can be generated.

Description

technical field [0001] The present invention relates to a text image generation method, in particular to an end-to-end image generation method capable of maintaining the text style in a complex background environment, which is mainly oriented to picture translation scenarios, and the target text is drawn according to the source style, and the background texture is preserved And the style of the original text (font, color, shape, etc.), to achieve high-fidelity replacement of the translated text. Background technique [0002] Style-consistent text-to-image generation aims to achieve high-fidelity text replacement, and has many practical applications, such as image translation, text detection and recognition tasks, movie poster text editing, etc. For image translation tasks, it can improve translation results and enhance user experience; for text detection and text recognition tasks, it can quickly expand data for specific scenarios; for designers, it can quickly design poster...

AI Technical Summary

Problems solved by technology

[0009] (1) It cannot handle scenes with more complex backgrounds, and artifacts are prone to occur in the process of reconstructing the background

[0010] (2) It is difficult to simulate complex deformation of text, such as perspective deformation, curved text, etc., and it may be difficult to ensure semantic coherence with the global image after reverse correction

[0011] (3) Unable to fully capture the style of the original text, such as fonts, shapes, shadows and other effects

Images