Audio analysis is said to offer better protection than optical methods
When it comes to the automatic detection and filtering of unwanted pornographic video content on the Internet, a considerable number of different software solutions are available to users today. The systems offered, which are based entirely on the optical analysis of the content, have their weak points and do not always work perfectly. In order to be able to guarantee better protection against online sex videos, Korean computer scientists have now presented a novel approach that does not rely on image recognition but on speech recognition to block dirty content.
No one hundred percent protection
"Security solutions that filter out websites and content with pornographic content are in great demand, especially in the area of so-called 'parental control'", explains Martin Penzes, Technical Director of ESET Austria http://www.eset.at , in conversation with pressetext. Although a large part of it already works very well in practice, there is still no one hundred percent protection. "I can already imagine that a filter that works by audio analysis can block sex videos more effectively," says Penzes.
In order to achieve the best possible protective effect, the ESET-Esperte advises a combination of image and audio analysis systems: "Both approaches have their advantages and disadvantages." While the visual analysis often has problems distinguishing sexual representations from conventional fashion images, where a lot of skin can also be seen, with the latter variant, no still images can be recognized and filtered.
93 percent recognized correctly
Despite the obvious weaknesses, Myung Jong Kim and Hoirin Kim, two computer engineers at the Speech Recognition Technology Lab, hold http://srtlab.kaist.ac.kr/ at the Korea Advanced Institute of Science and Technology (KAIST) continues to develop its audio-based porn filter system. As they report to the New Scientist in a recent article, the system they presented was able to correctly identify and block pornographic content in videos in 93 percent of the cases in a comprehensive test run.
(ck)
.