Content-Aware Image Filtering

Overview

Content-Aware Image Filtering uses AI to detect and blur NSFW (Not Safe For Work) images in real-time as students browse the web. When an explicit or inappropriate image is detected on a web page, it is automatically blurred before the student sees it — even on sites that are not blocked by your filtering policy.

This adds a visual safety layer on top of URL-based filtering. A website might be allowed by policy (e.g., a news site or social media), but individual images on that site can still be caught and obscured.

Prerequisites

• KyberGate Chrome extension v2.9.0+ installed on student devices
• Admin access to the KyberGate dashboard
• Extension permissions configured to allow content analysis

How It Works

1. As a web page loads, the Chrome extension scans images on the page using a lightweight on-device AI model
2. Each image is classified into categories: Safe, Suggestive, NSFW, Violence, Drugs/Alcohol
3. Images that exceed your configured threshold are blurred with a frosted glass effect
4. The student sees the blurred placeholder with a small shield icon indicating the image was filtered
5. Blurred images cannot be unblurred by the student
6. A log entry is created in the dashboard for review

Image analysis happens locally on the device — images are not sent to external servers for classification. This preserves student privacy and works even when the device is offline.

Steps

Step 1: Enable Image Filtering

Log in to the KyberGate dashboard → Settings → Filtering → Image Filtering. Toggle to Enabled.

Step 2: Configure Detection Categories

Choose which image categories to filter:

• NSFW / Explicit — nudity, pornography, sexually explicit content. Default: ON (recommended always on)
• Suggestive — provocative but not explicit content (e.g., swimwear ads, suggestive poses). Default: OFF for high school, ON for elementary/middle
• Violence / Gore — graphic violence, blood, injury imagery. Default: ON
• Drugs / Alcohol — images depicting drug use, paraphernalia, alcohol consumption. Default: OFF
• Weapons — images of firearms, knives, and other weapons. Default: OFF

Step 3: Set Detection Threshold

The threshold controls how sensitive the detection is:

• High Sensitivity (0.3) — catches more images but may have more false positives (some benign images may be blurred)
• Medium Sensitivity (0.5) — balanced detection. Recommended for most schools.
• Low Sensitivity (0.7) — only blurs images with high confidence. Fewer false positives but may miss borderline content.

The threshold value represents the AI model's confidence level — images scored above the threshold are blurred.

Step 4: Configure Per-Grade Settings

Different grade levels may need different sensitivity:

• Elementary (K–5) — High sensitivity, all categories enabled
• Middle (6–8) — Medium sensitivity, NSFW + Violence enabled
• High School (9–12) — Medium sensitivity, NSFW enabled

Configure per-OU under Organizations → select sub-org → Image Filtering.

Step 5: Test the Filter

KyberGate provides a test page at dashboard.kybergate.com/test/image-filter that displays sample images at various detection levels. Use it to verify the filter is working on a student device before rolling out.

Step 6: Review Filtered Image Logs

Go to Reports → Image Filtering to see:

• Total images filtered over the selected period
• Breakdown by category (NSFW, Violence, etc.)
• Top websites triggering image filtering
• Per-student filtered image counts
• False positive reports

Step 7: Handle False Positives

If a legitimate educational image is being incorrectly blurred:

1. Find the entry in the Image Filtering report
2. Click Mark as False Positive
3. Choose to whitelist the specific image URL or the entire domain for image filtering
4. The image will no longer be blurred on future page loads

Tips

• Start with Medium sensitivity and adjust based on false positive rates
• NSFW filtering should always be enabled — it's the most important category
• Suggestive content filtering is useful for younger students but may over-filter for high schoolers (clothing ads, health content)
• Image filtering works on Google Image Search results, social media feeds, news sites, and any web page with images
• The on-device model is small (~5 MB) and does not impact browser performance
• Image filtering does not affect images in Google Docs, Slides, or other productivity apps — it only applies to web page images
• Review the "Top Websites" report to identify sites that frequently trigger filtering — you may want to block some of them entirely

FAQ

Q: Are filtered images sent to KyberGate's servers? A: No. All image analysis happens locally on the student's device using an on-device AI model. Images never leave the device. Only metadata (URL, category, confidence score) is logged.

Q: Does image filtering slow down browsing? A: Minimally. The AI model processes images in under 50ms each. On pages with many images, there may be a brief moment where images appear before being blurred. The blur effect is applied as images load.

Q: Can students right-click and save blurred images? A: No. The blur is applied at the rendering level. Saving or downloading the image results in a blurred file.

Q: Does this work on video thumbnails? A: Yes. Video thumbnails on YouTube, Vimeo, and other platforms are treated as images and filtered normally. It does not filter video content while playing.

Q: What about images in PDFs or downloaded files? A: Image filtering applies to web content only. Downloaded files and PDFs are not scanned.

Q: Can I see the original image that was blurred? A: Admins can view the original image in the Image Filtering report (behind a click-to-reveal toggle) for review purposes. This is restricted to admin accounts only.