Introduction

Impero Wellbeing subscribers get exclusive access to the viewed word capture system, that triggers full screen captures on devices when certain words are detected. These captures are then sent to Impero Backdrop where users with the appropriate access can perform a triage against them with actions such raise a concern or mark as false positive. For more details on capture triage, click on the following link.


Users can deploy a Backdrop device client application. This application connects to Backdrop and downloads keyword lists covering a wide variety of e-safety categories including, for example, weapons and violence, bullying, or radicalisation. Users can also create their own collection of custom keyword lists. For more details on keyword lists, click on the following link.


When a keyword is detected on-screen capture is silently taken and sent to Backdrop listing Who, When, and What.

Who – The identified user on the device when the capture was triggered.

When – The time and date the capture was triggered.

What – The keyword term that triggered the capture.


The screen capture process takes 5 screenshots, 1 per second from the point the capture was triggered. These additional images can be viewed from the single capture view in Backdrop.


Device Client Behaviour

All the device client applications provide the functions listed above and it is important to understand how each platform delivers that functionality. We strive for parity across platforms but sometimes restrictions on certain device types mean that a different approach is necessary.


Windows

Captures viewed words across all applications and across all device displays. Scans the device display once every 50 seconds; this allows time for the previous scan to complete.


Once a capture is taken for a viewed word, that word is not going to be recaptured for the next 3 minutes, (scanning still happens during this period). This is to prevent Backdrop from being put under excessive load.


The scans are conducted with OCR (Optical Character Recognition) and rely on the ‘tessdata’ folder found in the install location. Written word captures runs in parallel to viewed word captures.


Prior to using Wellbeing, make sure that you are aware of the known limitations:

  • Scans for words only in English currently.
  • Does not work as well with applications that use the ‘dark mode’ theme.
  • Looks for single words only, ‘hiding bruise’ and other phrases do not trigger.


ChromeOS

Scans the HTML content of loaded webpages. A subsequent scan occurs when the HTML content of the webpages change. Written word captures runs in parallel to viewed word captures.

Prior to using Impero Wellbeing, make sure that you are aware of the known limitations:

  • Scans for words only in English. (should scan any word providing the characterisation matches, so accents in french for example)
  • Looks for single words only, ‘hiding bruise’ and other phrases do not trigger. (should detect words with spaces, really needs testing to confirm)


Android

Web pages are also examined via HTML scraping – any text displayed on a webpage that contains a keyword triggers a screenshot of the webpage. Different text areas and custom text controls are not currently observed. It only generates a match on the first occurrence of the word in order to avoid spamming the same thing repeatedly. 

Known limitations:  

Capture may not contain keywords if further down the page. 
Can only scan content in Impero web browser (this does not include in app menu/features, web pages only).


iOS

Once the content of a webpage is fully loaded, the application parses and scans throughout the HTML, ignoring text that forms part of the syntax/grammar of the HTML itself, and builds up a representation of the page in plain text. It then scans through the text character by character looking for matches against the list of concern words. It only generates a match on the first occurrence of the word in order to avoid spamming the same thing repeatedly.

Scans for words only in English. (should scan any word providing the characterisation matches, so accents in french for example)

  • Looks for single words only, ‘hiding bruise’ and other phrases do not trigger. (should detect words with spaces)


Sometimes, it cannot identify words in a text that are added to the page after the initial load.