Shadow Edge Orientation Map

The image is split into a grid. In each block the tool finds pixels with a strong luminance gradient whose dark side, sampled 3 px back along that gradient, falls below luminance 100 — a shadow edge — and averages their orientation. A block needs at least 10 such edges before it reports anything. The result is the orientation of the shadow boundaries, which is not the same thing as the direction of the light: the gradient at an edge is the boundary normal, and the normals around a cast shadow's outline point every way regardless of where the sun is. So this is a map to look at, not a test to pass. For compression-level forgery detection, combine with the ELA Scanner or JPEG Ghost Scanner.

Each arrow carries one block's mean shadow-edge orientation. The colour ramp runs blue to amber and shows how far that block sits from the dominant orientation of this frame, scaled against the largest departure in the same image — so one block is always at the amber end, in every image, and the colours mean nothing across two different images. Amber is a place worth looking at, not a finding. Gray blocks found fewer than 10 qualifying edges, which is common in bright or flat areas. Images larger than 2048 px on the long side are analyzed at 2048 px, and the size actually used is printed under the overlay. For the same image seen as raw brightness gradients rather than shadow edges, check the Luminance Gradient tool.

This page used to print a consistency percentage and a three-band verdict, red included. It was measured against the samples and against a synthetic control in which every cast shadow points the same way, and it did not separate them: the clean sample and the manipulated one scored within a point of each other, the synthetic single-light control landed in the middle band, and turning every shadow 180 degrees moved it by two points. A number that cannot tell those apart has no business accusing anyone, so the score and the verdict were removed rather than re-thresholded. What is left is the overlay. Real constraints remain either way — mixed indoor lighting, curved surfaces, reflections and flat or overexposed regions all change what the edges look like. Cross-reference with Noise Analysis, Chromatic Aberration, and the Authenticity Checker for a comprehensive assessment.

Shadow edge orientation is one view among several. Combine it with other manipulation detectors. Check Clone Detection for copy-paste regions. Use Demosaicing Trace to check whether the sensor interpolation trace survives at all. Run the Forgery Heatmap for a combined probability overlay. Together these tools cover compression artifacts, sensor-level patterns, pixel-level cloning, and, here, where shadow boundaries sit. Explore all forensic and utility tools at Scanly's full toolkit.