Brand Compliance Engine

The hard part

A logo in a real marketing asset is rarely the logo in the brand book. It is small, angled, half covered, or lit wrong. Raw detection accuracy collapses on those cases, so the gains came from the image processing around the model rather than from the model alone.

The problem

Brand teams need to know whether marketing assets actually follow the brand book, but logos in real assets appear at arbitrary scale, angle, and lighting, often partially covered.

The approach

I trained a YOLO based detector and wrapped it in image processing that normalizes scale, lighting, and viewpoint before inference, then layered typography and color checks on top of detection.

What it does

• YOLO based logo detection
• Pre processing for scale, lighting, and viewpoint normalization
• Robustness to partial occlusion
• Typography consistency validation
• Brand color matching across assets

Architecture

A Python pipeline where OpenCV pre processing feeds a YOLO detector, and detected regions pass through typography and color validation stages before a compliance verdict is produced.

Outcome

Extends a detection model into an actual brand compliance verdict, covering not just whether the logo is present but whether the surrounding type and color follow the brand system.