2015.02.09
NTT Comware Releases Beta Version of Automatic Filtering System to Protect Users from Viewing Inappropriate Online Content
- Deep Learning Technology to Simulate Human Cognition -
NTT Comware Corporation (Head office: Minato-ku, Tokyo; President: Shinobu Umino; hereinafter, NTT Comware) has successfully developed a system to recognize inappropriate online content, including pornographic images, at high resolution from SNS (Social Networking Services) or other websites.
This system contributes to making web administratorsf monitoring activities efficient; activities which used to require a huge amount of man-hours and cost through visual inspections. The service is easy and quick to use, as well as handy and flexible thanks to options to customize the program to meet various customer needs.
NTT Comware and Datasection Incorporated (Head office: Shibuya-ku, Tokyo; President: Hiroshi Sawa; hereinafter, Datasection) jointly launched the service on February 9, 2015, aiming for commercialization.
1.Background
The number of SNS users in Japan has been growing year by year. Along with this, the astonishing growth of all digital content, not only texts but also images or movies, continues to gather pace. Meanwhile, inappropriate content such as prank pictures on SNS has become a social issue and is also increasing. To monitor SNS to prevent such image content, it requires companies to conduct a visual inspection by human eyes, as it is considered difficult to rely on computer systems.Under such circumstances, NTT Comware has developed an automatic monitoring system to detect inappropriate online content in order to further contribute to a safe and secure information-oriented society in a modern world where children can easily access the Internet. The system applies a new method to recognize images using a standard to simulate human senses and filters online content in a more flexible and practical manner. For the system, "Deep Learning*1", a machine learning technology, is used to achieve high precision to recognize images.
2. Features of NTT Comwarefs Inappropriate Content Filtering Service (beta version)
The filtering system automatically recognizes and deletes inappropriate online content such as pornographic images.
(1) Optimum ability to rate gthe range", simulating human senses
NTT Comwarefs image recognition system evaluates the content of a target image with a numerical value called Score. The evaluation system categorizes restriction levels in a specific manner and is not a simple judgment of gappropriate" or ginappropriate". For instance, pornographic content for adults is classified into the restriction levels "G", "PG12", "R15+" or "R18+" in Japan, which are the rating standards applied and managed by the Film Classification and Rating Committee in Japan. The service is also flexible since the rating standard for classification can be set by the service operator. Therefore, the service operator can change the restriction levels of "inappropriate" content to remove such content. (Patent pending)
Such a score-based classification is optimum, and it is useful to have a grating" system not only for content for adults but also for various other purposes.
(2) Quick start
In general, machine learning needs vast amounts of data for learning and parameter tuning of the system to achieve a certain level of recognition precision. NTT Comwarefs image recognition system can create a high-precision search engine in a short term with less data for machine learning. Using the service saves a lot of time. (Patent pending)
(3) Custom-made service
NTT Comwarefs image recognition system is also available for custom-made designs to meet various customer needs. It is not only for automatic filtering of inappropriate online content but also for other purposes, developed with the know-how accumulated through the companyfs R&D activities in machine learning methods.
3. Future Scenario
Through the beta system service, we explore customer needs and services appropriate for such needs and are validating the system aiming for commercialization of the filtering service for launch in the first half of FY2015. The service may also be available for various specific purposes upon requests of users, other than for filtering of inappropriate content, to explore the potential of the service.The range of technology is also expected to extend o various fields including structural health monitoring of infrastructure such as bridges and tunnels and detection of unlawful activities using surveillance cameras on a real time basis to help create a safe and secure society in the future.
** Names of products and services are either trademarks or registered trademarks of each corporation.
