Image recognition AI that detects inappropriate images such as adult images, imitation product images, and illegal usage of logos and characters.

"ROKA SOLUTION" is an AI-type Image Recognition System that utilized deep learning, developed by E-guardian in cooperation with Tokyo University, that has world-class image recognition machine learning system.
Implements image detection rate with high-accuracy of 99.5% through the data that helps the AI to learn that utilized data acquired from extensive experience in post monitoring. It can detect inappropriate images such as adult images, imitation product images, etc.

System Flow Chart

Teacher data making that enables high detection rate

Our knowledge that have been accumulated years and years through the monitoring enables to make “teacher data” that will be required as a criteria to raise the detection rate for certain images. Images will be justified either “in” or “gray” or “out” and only the gray ones will be manually checked by the operators. In the case the judgement differs between the filtering and the visual checking, teacher data will be auto-corrected that the more we operate this system the more accurate judgement will be.

Filtering that can be used in various scenes such as EC, advertisement distribution, etc.

By filtering adult images and inappropriate images that violate social standards with a detection rate of 99.5%, it can be used not only to support healthy site management, but also in checking of advertisement distribution destination which is becoming a social problem nowadays. Also, in the EC field, the checking of fake products on CtoC flea market application sites, the recommend function which extracts image data that matches the taste of users, the suggestion of recommendation coordinate by AI in cooperation with "face authentication" system, etc. can also be applied.

Frequently Asked Questions

In trouble since imitation of products handled in our company are being sold in auctions

It is possible to detect perfect-match images and processed images by letting AI learn about the images that are illegally displayed in the past as a "teacher data". In addition, since the checked images are scored respectively, extraction of similar images can be easily performed by confirming the image with high score even if Ai judgement is not NG.

For Inquiries regarding Services and Quotations