M6Image forensicsMicroscopyLayer 1 (Deterministic)

SEM Noise Profile

Checks the sensor-noise profile of an electron micrograph. A real SEM or TEM image carries shot noise everywhere at a roughly uniform level because the beam scans the field with the same statistics throughout. The indicator flags a noise level that varies between regions (a composite), a near-total absence of noise (synthetic or denoised), and an implausibly high signal-to-noise ratio.

Technical description

Subtracts a 5 by 5 median-filtered copy to isolate a noise residual, then estimates the noise level robustly with the median absolute deviation, sigma = 1.4826 * median(|r - median(r)|), globally and on 16 by 16 blocks, so specimen edges left in the residual do not inflate it. Three signals score: the coefficient of variation of the block levels (0.3 to 0.7 scales to 3.0) for spatial inconsistency, a global level below 2.0 adding 3.0 for absent noise, and a signal-to-noise ratio mean/level above 50 adding up to 2.0. The contributions are summed and capped at 5.0.

How it works

Layer 1 (deterministic): extracts a median-filter noise residual, estimates the robust noise level per block and globally, and scores three departures from a uniform single scan: a coefficient of variation of block levels above 0.3 (inconsistent noise), a global level below 2.0 (absent noise), and a signal-to-noise ratio above 50. Blocks more than two standard deviations from the mean level are reported with bounding boxes. The contributions are summed and capped at 5.0.

Why this matters

Electron microscopy is shot-noise-limited: the signal is built from counted electrons, so every real micrograph carries Poisson noise whose presence and spatial uniformity are physical facts of the instrument, set by the dwell time, beam current, and detector that stay constant across one raster scan. A region spliced from another capture brings a different noise level, and a synthetic or heavily denoised region brings too little noise, so the noise profile exposes composites and fabrications that the eye cannot see.

Score thresholds

0-1: Noise is present and spatially uniform, consistent with a single electron-microscopy scan
2-3: One anomaly: a noise level that varies across the frame, or an unusually high signal-to-noise ratio
4-5: Noise is nearly absent or strongly inconsistent across regions, consistent with a synthetic image or a composite of different sources

Limitations

The median-residual estimate is robust but pervasive fine texture can still raise the apparent level, and a low-magnification image with large flat regions can read as suspiciously smooth. Dwell time, detector gain, and frame averaging change the noise level legitimately, so a frame-averaged acquisition can trip the absent-noise or high-SNR checks. The SNR is a coarse proxy that depends on field brightness. The 16-pixel block bounds resolution. Whether noise scales with intensity (the photon-transfer question) is indicator M5, and the general wavelet-domain noise-consistency screen for any image is indicator I3; M6 specialises in the spatial noise profile of electron microscopy plus the absence-of-noise and high-SNR signatures I3 does not score.