Photo Censor/Blur Tool

About Photo Censor/Blur Tool

How to Use the Tool

Step	Action
1. Upload image	Click the upload area or drag and drop a file
2. Choose blur mode	Select Gaussian Blur (smooth) or Pixelate (mosaic blocks)
3. Set strength	Adjust the slider to control how heavily the content is obscured
4. Draw regions	Click and drag on the image to draw rectangles over sensitive areas
5. Adjust or remove	Click X on any region to remove it, or draw additional regions
6. Download	The effect is permanently baked into the output image

Blur vs Pixelate

Mode	How It Looks	How It Works	Best For
Gaussian Blur	Smooth, out-of-focus appearance	Averages nearby pixels using a Gaussian function, creating a soft blend	Faces, natural images, when you want a subtle look
Pixelate	Large coloured blocks (mosaic)	Divides the region into blocks and fills each with the average colour	Text, licence plates, classic censoring look, screenshots

Both modes permanently modify the pixel data in the output image. The original content underneath the blurred or pixelated region cannot be recovered from the downloaded file.

What to Censor: Common Scenarios

Scenario	What to Blur	Recommended Mode	Strength
Sharing screenshots with personal data	Email addresses, phone numbers, account numbers	Pixelate	High - text must be completely unreadable
Posting photos with faces	Faces of people who did not consent	Blur	Medium to high - features should be unrecognisable
Sharing car photos online	Licence plates	Pixelate or blur	High
Bug reports with sensitive data	API keys, passwords, tokens visible in screenshots	Pixelate	Maximum - no characters should be guessable
Real estate photos	Neighbour's property, house numbers	Blur	Medium
Medical or legal documents	Patient/client names, dates of birth, case numbers	Pixelate	Maximum

Security Considerations

Concern	Is It Safe?	Details
Can the blur be reversed?	No	The download contains only the blurred/pixelated pixel data. Original pixels are discarded
Is the image uploaded anywhere?	No	All processing happens in your browser using the Canvas API. Nothing is sent to a server
Does EXIF data remain?	No	Canvas export strips all metadata including EXIF, GPS, and camera information
Can AI reconstruct the text?	At high blur/pixelation, no	Use maximum strength for text - low strength may still allow AI-assisted guessing of characters
Is the original file modified?	No	The original file is untouched. You download a new, separate image

Important: use maximum strength for text censoring. Research has shown that lightly pixelated text can sometimes be reconstructed, especially for short strings like passwords or licence plates. When in doubt, censor more aggressively.

Blur Strength Guide

Strength Level	Effect	Safe For Censoring?
Low (10-20%)	Slight softening, details still partially visible	No - details can often be guessed or reconstructed
Medium (40-60%)	Details obscured, shapes still recognisable	OK for faces and large objects, not for text
High (70-90%)	Content unrecognisable, only colours remain	Yes - safe for most content
Maximum (100%)	Completely obscured, no recognisable features	Yes - recommended for text, numbers, and sensitive data

Why Light Pixelation Is Not Safe for Text

Light pixelation can be reversed by matching blocks against a reference alphabet rendered in the same font. In 2020, a security researcher publishing under the handle spipm released Depix, an open-source proof of concept that recovers plaintext from screenshots pixelated with common tools like Greenshot and GIMP. It works because linear box-blur filters are deterministic - the same text rendered in the same font at the same block size always produces the same averaged colours, so a brute-force lookup against a De Bruijn reference sheet recovers the original characters.

JUMPSEC Labs published results in 2022 replicating the attack against common monospace fonts such as Consolas, Monaco, and Courier New. Short, high-entropy strings like passwords and MFA codes were the easiest to recover. Proportional fonts and heavy compression reduce success rates but do not make redaction safe. In 2021 a US cybersecurity researcher used unblur techniques on a Washington Post report to reconstruct blurred email addresses in court documents, confirming the attack generalises beyond lab conditions.

Worked example of the risk: a screenshot of a six-digit authenticator code pixelated at 8 px blocks in Consolas has roughly 10⁶ possible values. Depix can evaluate all of them in under a second on a laptop, so the "censored" code is effectively plaintext for an attacker. Raising the block size to 32-40 px and switching to a proportional font roughly halves the recovery rate, but the only provably safe option for sensitive text is to paint a solid opaque rectangle over it. This tool achieves that at maximum strength because the block size exceeds the character height - use the slider at 40 or above when covering codes, keys, or short identifiers.

What Data Can Leak From a Shared Photo?

Shared photos leak far more than the visible subject. A typical JPEG from a modern smartphone carries EXIF metadata that records the exact GPS coordinates, altitude, timestamp, camera model, serial number, and lens used. According to the UK Information Commissioner's Office (ICO), accidental disclosure through photo metadata remains a recurring personal-data breach category, and mainstream social platforms strip only a subset of this information on upload. Facebook, Instagram, and WhatsApp remove GPS and camera serial on public posts but retain them internally; direct messages on Telegram and Signal preserve the full EXIF block when sent as a "document" rather than a compressed image.

Data Field	What It Reveals	Stripped by This Tool?
GPS latitude/longitude	Exact location to within a few metres	Yes - Canvas export discards EXIF
Camera serial number	Links all photos from the same device	Yes
Timestamp	Exact date and time of capture	Yes
Device model	Phone or camera used	Yes
Editing history (XMP)	Software used and prior edits	Yes
Visible faces, plates, text	Direct personal data	Only if you draw a region over them

The Canvas API re-encodes the image from pixel data, so the downloaded file contains only the visible pixels - nothing else. This behaviour is standard across all modern browsers per the HTML Living Standard. For extra safety, you can also strip metadata from untouched images using a dedicated image compressor pass, which goes through the same Canvas pipeline.

How Gaussian Blur and Pixelation Work Internally

Gaussian blur convolves each pixel with neighbours weighted by a 2D Gaussian distribution, producing a smooth falloff controlled by the radius sigma. Pixelation divides the region into N×N blocks and replaces every pixel in a block with the block's average colour, which is mathematically equivalent to a box-filter downsample followed by a nearest-neighbour upsample. The tool uses the browser's native CanvasRenderingContext2D.filter = "blur(Npx)" for Gaussian mode and a temporary downsample canvas for pixelation, both of which are GPU-accelerated in Chrome, Firefox, and Safari.

Rule of thumb: to defeat pixel-matching recovery, the effective block size should be larger than the character height being covered. For 14 pt screen text that is roughly 20 px at typical DPI, so a pixel block of 24-32 px is the minimum. The strength slider in this tool maps directly to pixel-block size at a 1:2 ratio, so set it to 40 or above for any text containing passwords, tokens, medical numbers, or financial details. Faces are more forgiving because human features carry lower high-frequency information than text - a strength of 20-30 is usually enough to defeat both human and automated re-identification at social-media resolutions.

Best Practices for Sharing Redacted Images

Before publishing any censored image, apply these four rules. They are drawn from guidance by the UK ICO, the US NIST Privacy Framework, and published redaction-failure case studies.

Rule	Why It Matters
Use a solid opaque rectangle (max strength pixelation) for text	Depix and similar tools recover low-strength pixelated text with 70-90% accuracy on common fonts
Redact in a rasterised export, never a PDF annotation layer	PDF annotations can be removed; only flattened pixels are safe
Check the scroll area and reflections	Address bars, taskbars, notification pop-ups, window reflections, and screen glare all leak data
Recheck after resizing	Downsizing can make previously unreadable text legible if the original was over-blurred
Strip metadata	Canvas export (this tool) discards EXIF and GPS. Other tools may preserve it

For full-image artistic softening rather than targeted censoring, try the image filters tool. For removing unwanted areas entirely rather than obscuring them, the image cropper is more appropriate - cropped pixels are gone, not blurred. If you want to shrink the censored output for email or a web upload, run it through the image compressor afterwards. All three tools work entirely in your browser and never transmit your image.

Sources

• Depix - Pixelated Text Recovery PoC (GitHub)
• JUMPSEC Labs - Can Depix Deobfuscate Your Data?
• BleepingComputer - Researcher Reverses Redaction
• UK ICO - Reporting a Personal Data Breach
• NIST Privacy Framework
• HTML Living Standard - Canvas API
• MDN - CanvasRenderingContext2D.filter