Method and device for generating smeared images, storage medium and terminal equipment

Abstract

The embodiment of the invention discloses a method and device for generating a smear image, a storage medium and a terminal device, which are applied in the technical field of information processing based on artificial intelligence. The smear image generation device can automatically detect the area information of the character area of ​​the original image, generate multiple points in the corresponding character area, and determine the order between the multiple points, and then draw lines between the multiple points according to the determined order, Get the smudge image, which automatically smudges the traces in the character area of ​​the original image. In practical applications, in the process of training the smeared image restoration model, the corresponding multi-style smeared images can be automatically generated without manually marking the original image containing characters, which meets the training sample size requirement in the training process. Requirements and multi-style requirements, while simplifying the process of obtaining training samples.

Description

technical field [0001] The invention relates to the technical field of information processing based on artificial intelligence, in particular to a method and device for generating a smear image, a storage medium, and a terminal device. Background technique [0002] In the process of text recognition, if the text part in the text image is smeared, the current text recognition algorithm cannot recognize the text on the smeared part. In order for the text recognition algorithm to recognize the smeared image, it is necessary to train a smear recovery model. In this way, after the text smear image is input to the smear restoration model, the restored image can be output. [0003] In the prior art, when training the smear restoration model, a large number of non-smeared original images and smeared image pairs are required as training data. Generally, manual labeling is used. First, a large number of non-smeared text images are collected, and then manually on the text images. Add ...

AI Technical Summary

Problems solved by technology

[0002] In the process of text recognition, if the text part in the text image is smeared, the current text recognition algorithm cannot recognize the text on the smeared part. In order for the text recognition algorithm to recognize the smeared image, it is necessary to train a smear restoration model. In this way, after the text smear image is input to the smear restoration model, the restored image can be output

[0003] In the prior art, when training the smear restoration model, a large number of non-smeared original images and smeared image pairs are required as training data. Generally, manual labeling is used. First, a large number of non-smeared text images are collected, and then manually on the text images. Add a variety of smears to get smeared images, and in order to make the smear restoration model have a good recovery ability for different types of smeared images, it is necessary to have a large variety of training data, including the complexity of smearing, location, Direction, color, etc. Therefore, manual labeling will consume a lot of time and labor costs

Images