Picture identification method and device, computer equipment and medium

Abstract

The embodiment of the invention discloses a picture identification method and device, computer equipment and a medium. The method comprises the following steps of: inputting the first picture uploadedby the user into a de-interference auto-encoder for preprocessing so as to filter interference in a first picture and output a second picture, wherein the de-interference auto-encoder is obtained bytraining at least two types of interference sample sets, and disturbance modes added in the different types of interference sample sets comprise at least two of noise, affine change, filtering fuzzification, brightness change and monochromization; And inputting the second picture into a picture yellowing identification model for identification. According to the embodiment of the invention, the pre-trained anti-interference auto-encoder is used for carrying out anti-interference preprocessing on the picture uploaded by the user, so that the problem that the filth picture is successfully uploaded to a network through the identification of a picture yellowing model by using an anti-sample technology is solved; Pictures processed by an adversarial sample technology can be identified, the filtering capability of the adversarial samples is improved, and the network environment is purified.

Description

technical field [0001] The embodiments of the present invention relate to Internet technologies, and in particular to a picture recognition method, device, computer equipment and media. Background technique [0002] Most Internet applications allow users to upload avatars, pictures, etc., and the content of the pictures is all-encompassing. However, the country has strict regulations on the content of pictures uploaded to the Internet, and it is forbidden to upload and share pornographic pictures, that is, pornographic pictures. Therefore, before the picture is successfully uploaded to the network, it must be tested to determine whether it is a picture that cannot be spread on the network such as an obscene picture. [0003] At present, some mainstream AI companies, cloud service vendors, and business security vendors all provide API services to detect whether pictures and avatars are obscene and pornographic pictures, referred to as pornography services. Usually, pornogra...

AI Technical Summary

Problems solved by technology

[0004] However, when some unscrupulous users (network hackers) are not sure about the pornography service model used by the pornography service, they generate disturbing pictures by adding disturbances to the uploaded pictures, and carry out black-box attacks on the pornography service model, trying to bypass Obscene and pornographic images cannot be recognized normally through pornography filtering, so that obscene and pornographic images can be published on the Internet, disrupting the order of the Internet

Images