Nirmion
Hilfe Finden Sie ein Werkzeug

Image Tools / Local studio

Image Noise Estimate Report

Estimate high-frequency luminance variation as the root mean square residual from a local three-by-three median and export JSON.

Runs locallyNo image upload to Nirmion

Step 1 / Source

Choose one image

JPEG · PNG · WebP · 20 MB

Drop your image here

Or browse for a JPEG, PNG, or WebP image that you own or are authorised to edit.

One file · up to 20 MB · maximum 50 megapixels

Step 2 / Output

Image Noise Estimate Report settings

This analysis uses one documented method so repeated runs on the same decoded pixels remain comparable.

Prepare the RMS local-residual JSON reportThe source is not overwritten. Analysis produces a separate RMS local-residual JSON report.
Keep this page open until the download begins.
01No server upload

The selected image is decoded and re-encoded inside this browser tab.

02Working copy released

After download, the preview URL, decoded bitmap, canvas, and selected source reference are cleared.

03Your download remains

Nirmion cannot remove the finished file from your device or browser download history.

Workflow guide

Use image noise estimate report with an inspectable review path

Each visible luminance is compared with its clamped three-by-three neighbourhood median; squared residuals are averaged and square-rooted. Compare the result with the untouched source before using it downstream.
A

What you need

Choose one authorised JPEG, PNG, or WebP still image up to 20 MB and 12 million decoded pixels. Animated frames are not processed.

B

Social-ready choices

Compare repeated captures of the same static scene and dimensions while treating the score only as one high-frequency variation signal.

C

What you receive

Review and download a separate RMS local-residual JSON report; the original file on your device remains unchanged.

Decision guide

What image noise estimate report measures or changes

Each visible luminance is compared with its clamped three-by-three neighbourhood median; squared residuals are averaged and square-rooted.

Texture, sharpening, compression, edges, demosaicing, scale, and resampling affect residuals, so the score cannot isolate sensor noise or grade quality. Browser-local processing avoids a Nirmion file upload, but it cannot remove copies stored elsewhere.

OpenCV image filtering tutorial

Useful checks for image noise estimate report

  1. Compare repeated captures of the same static scene and dimensions while treating the score only as one high-frequency variation signal.
  2. Create a bounded review result while retaining the original for comparison, correction, or a different method.
  3. Test representative flat areas, gradients, edges, text, transparency, and high-frequency texture before accepting the output.

Good to know

Important image boundaries

Why might a PNG not become smaller?
PNG is lossless and browsers do not expose a quality control for it. Choose WebP or JPEG when reducing file size matters more than exact lossless pixels.
Are animated images supported?
No. The workspace accepts still JPEG, PNG, and WebP files. Re-encoding an animated source would otherwise keep only one frame.
What does clearing the file mean?
Because processing is local, there is no Nirmion server copy to delete. The tab releases its working references after success; your downloaded output remains on your device.

Questions answered

Frequently asked questions

Does Image Noise Estimate Report upload my image?

No. Supported JPEG, PNG, and WebP decoding, pixel inspection, transformation, and export run in your current browser tab. Nirmion does not send the selected image to its API for this workflow. The result is a separate derivative or JSON report, while the original file on your device remains unchanged and should be retained for comparison.

How does Image Noise Estimate Report work?

Each visible luminance is compared with its clamped three-by-three neighbourhood median; squared residuals are averaged and square-rooted. Fully transparent samples are excluded from numerical reports, while visual effects preserve source alpha. The method does not infer subjects, restore missing detail, calibrate a display, or develop raw camera data. Compare edges, gradients, text, texture, and transparent areas with the original before relying on the result.

What are the limits of Image Noise Estimate Report?

Texture, sharpening, compression, edges, demosaicing, scale, and resampling affect residuals, so the score cannot isolate sensor noise or grade quality. The workspace accepts one supported still image up to 20 MB and caps this pixel-intensive operation at 12 million decoded pixels for predictable browser memory use. Results depend on browser decoding and encoding support. Animation, raw development, ICC calibration, layer editing, and batch processing are outside this tool's verified scope.