Interpreting Results

Image Analyzer ViS supports the classification of images for multiple threat categories. The scan results return a probability for each active category.

For example, if the categories Porn, Terrorism, and Weapons are activated, then a single image will return a probability for each category.


scan results: 


Porn : 0.2433

Terrorism : 99.67

Weapons : 89.486



The result returned is 0-100 and indicates the probability the image contains visual attributes common to the category. It is possible for an image to return high probabilities for multiple categories; in this example the image is of a terrorist holding a firearm and therefore returns high positive results for both Terrorism and Weapons categories.


Setting Thresholds

The higher the probability for a particular category the more likely the image composition matches that category. The default recommended threshold is 50%, this will provide high detection with false positives of approximately 1-2%. Increasing the threshold will decrease the false positives and detection rate. Lowering the threshold will increase the false positives and detection rate.

The table below provides a guideline for the level of false positive/detection rates at certain thresholds for the porn category.

Threshold Value

99%

95%

75%

50%

25%

5%

1%

False Positive Rate*

0.09%

0.20%

0.61%

1.67%

4.43%

9.40%

19.35%

Detection Rate*

67.66%

78.69%

87.08%

92.77%

97.00%

98.29%

99.30%


* Figures are indicative and will vary depending on image data and category. Dataset used to produce these figures was designed to challenge the technology.


If images that return a probability equal to or greater than 25% are considered pornographic then the detection rate will be approximately 97% with 4% false positives.

If images that return a probability equal to or greater than 75% are considered pornographic then the detection rate will be approximately 87% with 0.6% false positives.

Setting the optimum threshold will depend on the use case and the desired goal. It is recommended to start with a threshold of 50% and increase/decrease it as required.

 

 

Was this answer helpful?

 Print this Article

Also Read

Scan Image Data

This function is used when the image data is to be uploaded. It returns a JSON protocol that...

Scan Image URL

This function is used when the image data is to be downloaded from an image URL. It returns a...

Error Codes

  -1 = Username Empty or Invalid -2 = Password Empty or Invalid -3 = Not Enough Scan tokens...

Authentication

IA ViS requires the calling application to supply authentication credentials as ‘AUTHORIZATION’...