Micro-Texture Analysis

Examination of fine-grained texture patterns at small scales (patches of 32x32 pixels or smaller) to detect repetition, uniformity, or unnatural statistical properties.

AI models use pattern-matching from training data. When generating textures, they may repeat learned patterns, especially in areas like grass, fabric, or skin.

Autocorrelation measures self-similarity at different offsets. High secondary peaks indicate repeating patterns - a common AI artifact in texture generation.

Real textures have specific statistical distributions (skewness, kurtosis). AI textures often appear too Gaussian or overly symmetric - unnaturally "perfect" randomness.

Yes, AI skin often lacks natural pore variation or has unusually uniform texture. The detector analyzes local variance consistency to identify these artifacts.

AI-generated fabrics often show subtle tiling or periodic patterns. Real fabrics have manufacturing irregularities that create non-repeating variations.

By computing local variance in small patches across the image. Real photos have varying texture intensity; AI often has suspiciously consistent texture quality everywhere.

Smooth surfaces have less texture to analyze, reducing detection reliability. Other methods like lighting and edge analysis become more important for smooth subjects.

Some AI uses Perlin noise for variations, which has distinctive mathematical properties. The detector identifies these synthetic noise characteristics.

Micro-texture analysis is subtle and can be affected by compression and resizing. It provides supporting evidence rather than definitive detection.