Detecting pornography
< Previous help topic Next help topic >
The product allows you to detect pornography in image and video profiles.
In order to detect porn, select image or video profile of interest, right click it, choose Analyze images context menu item and then Detect porn. The analysis will start.
After the analysis the filter Porn images under Analysis results profile subnode in Case Explorer will be filled with the images with porn detected. The total number of images with porn will be indicated in the filter node:
43 porn images detected
Porn detection uses
neural networks in order to find faces. To the current state of art it is not possible
to create an algorithm which does such analysis 100% correctly, so one should expect
both false positives and false negatives. The following obstacles may prevent successful
porn detection:
- Poor lighting conditions
- Too few naked skin visible
- Various post-processing effects like converting colors to luminescent ones in image editor
Current implementation
has the following qualitative characteristics: 4,96% of false negatives and 16,31%
of false positives.
To help you to detect more suspicious images, the porn probability is assigned for each image. You can sort images list by the column Porn probability to have more suspicious images going first.
Please note, that for detection porn in a video file, you should extract key frames first.
