Mobile Guardian provides on-device image classification and blurring as part of its web filtering capabilities. When enabled, images displayed in the browser are analysed locally on the device and classified in real time. Images identified as inappropriate (e.g. explicit, violent, or otherwise harmful content) are automatically blurred before the student sees them.
Because classification happens on-device rather than in the cloud, it works with minimal latency and continues to function even when connectivity is intermittent. This feature is available across iOS/iPadOS, macOS, Android, Windows, and ChromeOS.
What You Will Learn
- How on-device image classification and blurring works
- How to enable the feature within a profile
- How classification categories are applied
- Platform-specific behaviour
Prerequisites
- Devices enrolled in Mobile Guardian
- Admin access to the Mobile Guardian Dashboard
- A profile (Baseline or Conditional) assigned to your devices
- The web filter enabled within the profile
How On-Device Image Classification Works
Traditional web filters evaluate URLs and page content at the domain or keyword level. Image classification adds a visual layer of protection by analysing the images themselves as they load.
| Step | What Happens |
| 1. Page loads | The student navigates to a web page in a filtered browser |
| 2. Images detected | The on-device classifier identifies image elements on the page |
| 3. Classification runs locally | Each image is evaluated against trained classification models directly on the device |
| 4. Decision made | Images are scored against categories (explicit, violence, drugs, etc.) |
| 5. Blur applied | Images exceeding the classification threshold are blurred in real time before rendering |
The student sees the page load normally, but inappropriate images appear blurred. The rest of the page content remains visible and accessible.
Note: On-device classification does not send images to an external server for analysis. All processing happens locally on the device, preserving student privacy.
Enabling Image Classification and Blurring
Step 1: Navigate to Profile Settings
- Log in to your Mobile Guardian Dashboard.
- Navigate to Profiles: Click on “Profiles” in the left-hand navigation panel.
- Edit the profile: Click the pencil icon under the “Actions” column for the Baseline or Conditional profile you want to configure.
Step 2: Open Safe Content Settings
- Under the “What” heading, click on “Safe Content”.
- Ensure “Enable Web-filter” is ticked. Image classification operates as part of the web filter.
Step 3: Enable Image Classification
- Under the “General” section of the Safe Content settings, locate the image classification option.
- Tick “Enable On-Device Image Classification”.
- When enabled, image blurring is applied automatically to images that are classified as inappropriate.
- Click “Save” to apply.
Classification Categories
The on-device classifier evaluates images against the following categories:
| Category | What It Detects |
| Explicit/Adult | Nudity, pornographic content, sexually suggestive imagery |
| Violence | Graphic violence, gore, weapons in violent context |
| Drugs | Drug paraphernalia, substance use imagery |
| Self-harm | Content depicting or promoting self-harm |
| Hate symbols | Extremist imagery, hate group symbols |
All categories are active when the feature is enabled. Classification thresholds are managed by Mobile Guardian and updated automatically to improve accuracy over time.
Note: Image classification uses machine learning models that are regularly updated. Occasional false positives (safe images incorrectly blurred) or false negatives (inappropriate images not caught) may occur. Use image classification alongside other filtering layers (DNS filtering, web filter categories, keyword filtering) for comprehensive protection.
Platform-Specific Behaviour
| Platform | Classification Runs On | Browser Coverage |
| iOS/iPadOS | Device (via Mobile Guardian app) | Mobile Guardian Safe Browser |
| macOS | Device (via Mobile Guardian app) | Mobile Guardian Safe Browser |
| Android | Device (via Mobile Guardian app) | Mobile Guardian Safe Browser |
| Windows | Device (via Mobile Guardian Edge extension) | Microsoft Edge with Mobile Guardian extension |
| ChromeOS | Device (via Mobile Guardian Chrome extension) | Google Chrome with Mobile Guardian extension |
Performance Considerations
- On-device classification requires processing power. On older or lower-specification devices, page load times may be slightly slower when the feature is active.
- Classification models are stored locally on the device. Model updates are delivered during routine Mobile Guardian syncs and do not require additional bandwidth during browsing.
- Battery impact is minimal under normal browsing conditions but may increase on devices with heavy image-rich browsing.
How Image Classification Interacts with Other Filters
Image classification operates as an additional layer within the Safe Content filtering stack:
- DNS filter (if enabled): Blocks the domain entirely before the page loads.
- Web filter categories: Evaluates the URL against category databases.
- Keyword filter: Scans page text content for flagged words.
- Image classification: Analyses images on the page and blurs those classified as inappropriate.
A page can pass DNS, URL, and keyword checks but still have individual images blurred by the classifier. This is useful for sites that are generally educational but may contain user-generated image content (e.g. search engines, forums, or wikis).
Verifying Image Classification Is Active
From the Mobile Guardian Dashboard
- Navigate to “Devices” > “All Devices”.
- Select a device to open the device details.
- Confirm the profile with image classification enabled is assigned and active.
From the Device
- On a test device, open the Mobile Guardian Safe Browser (or the managed browser with the MG extension).
- Navigate to a page containing images.
- Images classified as inappropriate will appear blurred, confirming the feature is active.
Troubleshooting
| Issue | Likely Cause | Resolution |
| Images not being blurred on any page | Image classification not enabled in the active profile | Verify “Enable On-Device Image Classification” is ticked under Safe Content > General |
| Images not blurred but web filter is working | Web filter is active but image classification is not enabled separately | Image classification must be explicitly enabled in addition to the web filter |
| Safe images being blurred (false positive) | Classification model flagging benign content | This may self-resolve with model updates. If persistent on a specific site, consider adding the domain to the web filter Allowed List |
| Feature not working on a specific platform | Browser not supported or extension not installed | Verify the Mobile Guardian app or browser extension is installed and active on the device |
| Page loading slowly with feature enabled | Device processing images in real time | Expected on lower-specification devices. If performance is significantly impacted, consider whether the device hardware is adequate |
| Blurring not working in non-MG browsers | Classification only runs within managed browsers | On iOS/macOS/Android: only the Safe Browser is covered. On Windows/ChromeOS: only the browser with the MG extension installed |
Best Practices
- Enable image classification on your Baseline profile so all devices have image-level protection regardless of which Conditional profile is active.
- Combine image classification with DNS-layer filtering and web filter categories for layered protection. No single layer catches everything.
- Inform students and staff that image blurring is active. Transparency about filtering builds trust and reduces support queries about why images appear blurred.
- Monitor for false positive reports from teachers. If a specific educational resource is being incorrectly flagged, address it through the web filter Allowed List rather than disabling classification entirely.
- Do not rely on image classification as your only filtering layer. It supplements URL and category-based filtering but does not replace it.
Please let us know if you found this helpful! Thanks for reading! 🙂