How does Strasty detect unwanted content?
Strasty uses several independent detection methods.
The results are combined before the program decides
whether content should be blocked.
1. Website address analysis
When you are using a supported web browser,
Strasty attempts to identify the address of the
currently opened page.
The domain can then be checked against the Strasty
database of known unwanted websites.
If a domain is already known to belong to a blocked
category, the application can react without having
to classify the page again.
2. Search query analysis
Search queries can also be analyzed for known
terms associated with blocked categories.
Search detection is handled separately from
ordinary text displayed on a webpage.
This distinction is important because the same
words may appear in news articles, conversations,
programming code or documentation without
indicating that the user is actually searching
for prohibited content.
3. OCR text recognition
Strasty periodically analyzes the visible screen
using OCR (Optical Character Recognition).
OCR converts visible text into text that the
program can compare with local stop-word lists.
Different lists are used for different categories,
including gambling, adult content and trading.
Finding a stop word does not necessarily mean that
the screen will immediately be blocked.
A word such as a casino-related term may appear in
a chat, spreadsheet, source code, news article or
technical documentation. Strasty therefore uses
additional signals to reduce false detections.
4. AI image analysis
Strasty also analyzes the visual content of the
screen using machine-learning models running
locally on the computer.
The models attempt to determine whether the screen
resembles normal content or one of the protected
categories.
The classifier can evaluate categories such as:
- normal content;
- pornography;
- nudity and sexual content;
- casino content;
- sports betting;
- forex;
- stock trading;
- cryptocurrency exchanges.
The AI model produces a confidence score.
Strasty uses confidence thresholds before
considering an AI result strong enough to trigger
protection.
5. Combining text and visual signals
One of the most important parts of Strasty is the
combination of different detection methods.
For example, OCR may detect a gambling-related
word on the screen. If the visual classifier sees
an ordinary application, source-code editor or
another normal screen, the word alone may not be
sufficient to trigger blocking.
If both the detected text and the visual analysis
indicate the same prohibited category, the
confidence that the content should be blocked is
significantly higher.
Strong visual detection can also trigger protection
independently when the AI model is sufficiently
confident.
6. Whitelist
Strasty also maintains exceptions that help prevent
known safe content from triggering unnecessary
blocking.
Together, these mechanisms allow the program to
make a decision using multiple signals instead of
simply blocking every page containing a particular
word.