Deterministic Visual Analysis for Out-of-Home Creative
The Scientific Foundations of Ad Corrector
Ad Corrector evaluates out-of-home (OOH) creative using deterministic image analysis and rule-based scoring logic. For the same image, text inputs, viewing speed, distance, and scoring mode, the output remains identical.
Modeling Statement: All calculations are explicit, weighted, and bounded. There is no adaptive training layer and no generative reasoning.
I. Deterministic System Architecture
Ad Corrector operates client-side for consistent, repeatable processing. The system utilizes:
- Canvas-based pixel sampling for layout and luminance analysis.
- OCR text detection via Tesseract.js to detect text regions and estimate bounding boxes.
- Rule-based scoring formulas and fixed weighting models.
- Explicit clamping and rounding logic to ensure numeric consistency.
Each component score is calculated within defined numeric limits, and composite scores are derived from weighted averages. Given identical inputs, the results do not change.
II. Relative Luminance and Contrast
Visibility depends on separation between elements. Ad Corrector samples pixel values and calculates relative luminance (L) using the standard photopic weighting model:
L = 0.2126R + 0.7152G + 0.0722B
This reflects human sensitivity to green wavelengths relative to red and blue. Contrast scoring is derived from luminance differences across sampled regions, using the same luminance weighting structure referenced in ISO/IEC 40500 for luminance and contrast evaluation.
III. Viewing Distance & Structural Resolvability
OOH creative is viewed at distance and often at speed. The system uses viewing distance and layout proportion signals informed by geometric visual angle relationships:
θ = 2 arctan(h / 2d)
This provides a physics-informed approximation of resolvability under defined viewing assumptions, using layout proportion signals rather than physical measurements.
IV & V. Motion, Exposure & Text Detection
Motion Adjustments
OOH creative is experienced under limited exposure time. Ad Corrector adjusts clarity and CTA scoring based on viewing speed, distance, element edge proximity, and spatial isolation. These adjustments estimate how motion and limited exposure affect clarity. All modifiers are deterministic and applied consistently.
Structural Signals via OCR
The system uses OCR to identify bounding boxes, support autofill, evaluate CTA placement, and analyze word counts. Text must meet confidence thresholds before influencing scoring. Adjustments include format-specific word density limits, action-oriented language recognition, and speed-based blur survivability.
The Persuasion Engine: Interpretation of Visual Signals
The Persuasion Engine transforms Step 1 component scores (Readability, Contrast, Clarity, Color Distribution, Composition, and CTA Effectiveness) into five persuasion signal metrics on a 0 to 10 scale. These values are calculated using fixed weighted relationships; they do not introduce new image analysis, but rather reinterpret existing structural signals.
Activation
Derived from weighted combinations of contrast, composition, and color distribution.
Motivation
Weighting depends on mode (incorporating CTA, clarity, and readability).
Identity
Combines clarity, composition, and color distribution.
Ease
Calculated by inverting a friction value derived from clarity, readability, and composition.
Memorability
Combines composition, color distribution, and contrast.
Persuasion Score: The final Persuasion Score is calculated as a weighted combination of the five metrics and normalized to a 0 to 100 scale. All transformations are deterministic.