Lighting Inconsistency Analysis

The foundational process of inferring the dominant light source vector from an object's shading pattern. By analyzing the intensity gradient across a surface—typically a face or a sphere-like object—algorithms solve for the Lambertian reflectance equation to compute the azimuth and elevation of the illuminant. A mismatch in the estimated light direction between two objects in the same scene is a primary indicator of compositing.

• Surface Normals: Calculated from a 3D model fit or photometric stereo.
• Cast Shadows: The angle of attached and cast shadows provides a geometric constraint on the light source position.
• Specular Highlights: The location of the brightest point on a glossy surface directly indicates the light vector.

This analysis estimates the chromaticity of the light source(s) illuminating the scene. The human visual system automatically corrects for illuminant color, but forensic algorithms can reverse this process. If one face exhibits a warm, tungsten-like illuminant while another shows a cool, daylight-balanced illuminant, the image is a composite. Techniques include analyzing the distribution of pixel colors in the inverse-intensity chromaticity space to separate body reflection from interface reflection.

The shape, size, and intensity distribution of specular highlights on the eyes, skin, or glossy objects must be physically consistent. A specular highlight mismatch occurs when the reflection in the left eye points to a different light source shape than the right eye, or when the highlight on a foreground object does not match the scene's dominant light. Deepfake models often generate plausible but physically incorrect highlights that fail a photometric consistency check.

Ambient occlusion describes the soft, darkening effect in crevices and areas where objects meet, caused by the occlusion of diffuse environmental light. In a composite image, a pasted object often lacks correct contact shadows where it meets the ground plane, or exhibits ambient occlusion inconsistent with the background's geometry. This is a critical cue for detecting object insertion, as generative models frequently fail to model this global illumination phenomenon accurately.

By transforming an image into the frequency domain via a Discrete Fourier Transform (DFT) or wavelet decomposition, analysts can detect inconsistencies in lighting that are invisible in the spatial domain. Splicing boundaries often introduce high-frequency anomalies at the seam. Furthermore, the lighting of a synthetic face generated by a neural network may exhibit a different frequency signature—such as a lack of natural high-frequency noise or unnatural banding—compared to the authentic background.

A highly robust forensic technique involves fitting a 3D Morphable Model to a face in the image. This process simultaneously solves for the face's 3D shape, texture, and the parameters of the scene's illumination using spherical harmonics. The recovered lighting parameters from a questioned face can then be compared to those from a reference face or the background. A statistically significant deviation in the estimated spherical harmonic coefficients is a powerful, quantifiable metric of forgery.

A forgery detection method that compares the position, shape, and intensity of specular reflections in the eyes or on surfaces. In authentic images, highlights from the same light source appear in geometrically consistent locations. In composites, highlights often reveal physically impossible light source positions.

• Compares corneal reflections between subjects in a scene
• Detects mismatched highlight shapes indicating different source environments
• Often combined with lighting direction estimation for 3D scene reconstruction

A forensic technique that fits a three-dimensional face model to a 2D image to estimate shape, texture, and illumination parameters. The recovered lighting coefficients from different faces in a scene must be consistent. Inconsistencies in the estimated spherical harmonics lighting between subjects indicate face-swapping or compositing.

• Recovers albedo, shape, and illumination separately
• Estimates full 3D light environment from a single face
• Detects impossible combinations of lighting direction and face geometry

A forensic method that re-saves an image at a known quality level and analyzes the difference from the original. Regions with different compression histories—such as spliced-in objects under different lighting—exhibit distinct error levels. This technique is particularly effective at identifying inconsistent shadow regions that have been composited from separate sources.

• Highlights areas with different JPEG compression histories
• Effective for detecting spliced objects with mismatched illumination
• Low computational cost for initial triage screening

A specialized forensic technique that examines cast shadows within a scene to geometrically constrain the 3D position of the light source. Shadows from different objects must converge to a consistent illuminant position. Divergent shadow directions or impossible penumbra characteristics indicate compositing from multiple source images with different lighting conditions.

• Uses vanishing points of shadow edges to triangulate light position
• Analyzes shadow softness for light source size estimation
• Detects physically impossible shadow-object relationships

A forensic method that estimates the illuminant color from different regions of an image. Each light source has a characteristic spectral power distribution that imparts a specific color cast. When spliced objects exhibit a different white balance signature than the background scene, it indicates they were captured under different illumination.

• Estimates scene illuminant from specular highlights and diffuse regions
• Detects mismatched color temperature between foreground and background
• Uses gray-world and white-patch assumptions for illuminant estimation

A video forensics method that evaluates the coherence of illumination across consecutive frames. In authentic footage, lighting changes smoothly with camera or subject motion. Manipulated videos often exhibit flickering illumination estimates or abrupt changes in light direction and color temperature that violate physical continuity.

• Tracks lighting parameters frame-by-frame for stability
• Detects abrupt illumination changes at edit boundaries
• Correlates lighting variation with expected scene motion