Content recognition and control systems - Visual content identification

Smilart has developed the simple and efficient recognition system of unsuitable visual content.

Analyzed content:

• Text;
• Photo;
• Video;
• Flash Clips.

Operation principle

First the system checks if the URL belongs to URL whitelists and blacklists. If the result is negative, the system starts content checking.

Text and included URLs are checked over whitelists and blacklists, while photo and video processing has several stages of intellectual analysis:

• Search of similar images or their parts in censorship and adult databases;
• Recognition and search of faces in person databases;
• Calculation of naked bust and body rate, and skin ratio in the image.

Note: the system can recognize inscriptions, logos, drawings, formulae, number plates, and symbols in uploaded images.

At each stage an image gets a threat level value (1-5):

• If an image is found in any of the databases, it gets the maximum threat level value ‘5’
• During the intellectual analysis the threat level is calculated on the basis of the values of the mentioned characteristics.

Check types:

• When the partial check is conducted, the system stops the operation and delivers the highest threat level if unsuitable words and URLs are found and any object belongs to any of the mentioned databases;
• The full check examines all the content.

Here you can see the operation algorithm of "Visual Content Identification":