Transform · Measure · Stress-test

Robustness Testing Studio

Stress-test image data and steganographic payloads against real-world transforms. Apply compression, resizing, cropping, blur, noise and tonal changes; measure image quality and probe whether embedded bits survive.

Important: browser simulations are controlled approximations. They do not reproduce Instagram's exact pipeline. Results apply only to this source image, selected transforms and browser encoder; verify with actual downloaded platform output.

01 · Input & probe payload

Use an image you own or have permission to test. Everything is processed in this tab.

Source image

Probe payloadA small test string is embedded into RGB least-significant bits before transformations. It is a fragile diagnostic probe, not a robust watermark or secret channel.

Embed LSB probe before transforms Repeatable noise pattern

Choose an image to start.

Original image preview

02 · Transform pipeline

Transforms are applied in this order: resize → crop → rotate → blur/tonal changes → noise → output encoding.

Resize (%)

Crop each edge (%)

Rotate

Output encoding

JPEG/WebP quality: 75%

Blur radius (px)

Brightness (%)

Contrast (%)

Noise intensity (0–40)

03 · Before / after & quality metrics

Dimension, size, MAE, PSNR, approximate SSIM and pixel histogram.

Original

Load an image

Transformed

Apply transforms

Dimensions—

Output size—

Size change—

MAE (0–255)—

PSNR—

Approx. SSIM—

Probe recovery status will appear after running a transform.

04 · Batch stress profiles

Run a suite of common transform profiles and compare recovery rates in one report.

Batch results will appear here.

05 · Reproducibility notes

Capture the conditions behind a test so results can be repeated and compared.

Experiment notes / platform observations

For actual social-platform validation: embed a probe, upload the image using the platform, download the processed media, then test the downloaded file separately. Never infer platform robustness from local simulations alone.

06 · Fragile vs robust embedding matrix

Embeds the same message two ways in the image from panel 01: plain RGB-LSB (fragile) and a keyed spread-spectrum watermark (blind detection, no original needed). Then runs both through ~25 real-world transforms and reports how many message bits survive.

Message (≤ 32 bytes)

Secret key

Strength α (1–12)

Cell size (px)

Load an image in panel 01 (just choose the file), then run. Larger α survives harsher edits but is more visible. Cell size trades resilience to scaling and compression against capacity. Working copy is capped at 1600 px.

Matrix appears here.

Detect the watermark in any image

Length (bytes)

Rescale to W×H (optional)

Uses the key, cell size and byte length above. If the picture's size differs it is rescaled to the size last embedded, or to the dimensions typed here.

07 · Compression quality sweep

JPEG or WebP at quality 95→10: how SSIM, LSB survival and spread-spectrum survival fall as quality drops.

Encoder

Results appear here.

08 · Compare two images

Pixel-exact difference report with an amplified heat map. Reveals invisible changes such as LSB embedding.

Image A (reference)

Image B

Heat-map gain ×

Results appear here.