In routine corporate documentation, warranty disputes, and civil contract workflows, users frequently photograph physical monitors or smartphone displays using their mobile devices rather than executing system-level screenshots. This pattern routinely introduces two substantial optical degradation modes: 1. Moiré Fringes: Curved, chromatic aliasing waves superimposed across the image raster, obscuring critical textual characters, micro-timestamps, and transaction hashes. 2. Projective Keystone Distortion: Off-axis sensor orientation creates non-orthogonal trapezoidal geometry where planar rectangles deform into asymmetric quadrilaterals.
The physical origin of moiré aliasing is governed by the Nyquist-Shannon sampling theorem. Physical display hardware features a fixed microscopic sub-pixel lattice. When the mobile camera's discrete sensor—itself arranged in a fixed periodic photo-site grid covered by a Bayer filter distributing green, red, and blue filters across photosites")—samples this emitter matrix, spatial beat frequencies emerge. The differential modulation between the camera's spatial sampling frequency and the monitor's emitter spacing manifests as macroscopic low-frequency interference waves.
Frequency-Domain Filtering via 2D-FFT
Conventional spatial smoothing filters (e.g., Gaussian kernels) fail because they attenuate high-frequency stroke details simultaneously with the interference patterns. The rigorous signal processing solution transitions the spatial image tensor into the spatial frequency domain via Two-Dimensional Fast Fourier Transform (2D-FFT).
Because the display's physical aperture grille is rigidly periodic, moiré energy concentrates into discrete, symmetric spectral impulses in the frequency plane. Applying adaptive notch attenuation masks allows surgical suppression of these periodic spikes without attenuating the broadband continuous spectrum embodying typography and character boundaries. Computing an Inverse Fast Fourier Transform (2D-IFFT) returns a pristine spatial image devoid of aliasing striations while preserving typographic edge definitions.
Orthogonal Rectification via Planar Homography
Following frequency domain filtering, perspective distortion is rectified via planar homography. By detecting the four bounding vertices of the display perimeter, a $3 \times 3$ projective transformation matrix $\mathbf{H}$ is solved:
$$\mathbf{x}' = \mathbf{H} \mathbf{x}$$
Applying bilinear back-projection reconstructs a geometrically true orthogonal document plane. Executing this pipeline client-side inside the browser sandbox guarantees that sensitive contracts, financial statements, and telemetry records remain completely private.