Convex character bank card number image synthesis method, recognition method and device

Abstract

The invention provides a convex character bank card number image synthesis method, a recognition method and device, and relates to the technical field of image synthesis. The method comprises the following steps: collecting a resource file required by a to-be-synthesized non-convex character bank card number image, wherein the resource file comprises a bank card number background image, arrangement rule information of bank card numbers and characters of the bank card numbers; synthesizing non-convex-font bank card number images based on the resource file and a non-convex-font bank card image synthesis algorithm, inputting the non-convex-font bank card number images into the trained annular generative adversarial network model, and synthesizing convex-font bank card number images corresponding to the non-convex-font bank card number images. The synthesized convex-character bank card number image obtained by adopting the method is relatively high in texture quality, so that the accuracyof recognizing the bank card number by the recognition model can be improved by adopting the synthesized convex-character bank card number image to train the recognition model.

Description

technical field [0001] The present application relates to the technical field of image synthesis, in particular to a synthesis method, recognition method and device for bank card number images in embossed fonts. Background technique [0002] In recent years, with the continuous development of Optical Character Recognition (OCR) technology, it has become possible to recognize text information in various certificate photos through OCR technology. Since training requires a large number of pictures, and the number of real pictures is difficult to obtain, and the number of acquisitions is limited, synthetic image samples are used to train the recognition model. [0003] At present, one method of synthesizing image samples is to use the Synthtext algorithm, but when synthesizing complex natural scene text images (especially images with complex textures, such as bank card images with convex font effects), the synthesis effect is often poor. Poor, resulting in low recognition model...

AI Technical Summary

Problems solved by technology

[0003] At present, one method of synthesizing image samples is to use the Synthtext algorithm, but when synthesizing complex natural scene text images (especially images with complex textures, such as bank card images with convex font effects), the synthesis effect is often poor. Poor, resulting in low recognition model accuracy

In addition, for synthesizing convex bank card number images, you can use Pix2Pix and other methods based on supervised learning to generate adversarial models. Although the three-dimensional and texture effects of bank card images are simulated very well, real bank card numbers need to be produced in the process of synthesizing training set samples. The image corresponds to a non-convex font image, and the bank card number in the real image needs to be manually labeled. When the training set is large, it often takes more time to manually label, which is costly.

Images