Overview
This started as a literature question -; which visual patterns, colours, and subjects reliably lift mood, and why -; and ended up as a Positive Emotion Image Scorer: software that takes an image and predicts how likely it is to trigger a positive emotional response, with the prediction grounded in actual neuroscience rather than vibes.
It's the visual sibling of the EEG and colored-noise explorations: same underlying curiosity about how a stimulus reshapes brain state, pointed at the eyes instead of the ears.
Background
The research note behind this pulls from neuroimaging, EEG, and cross-cultural studies spanning thousands of participants. A few findings stuck and shaped the scorer: positive stimuli light up reward centres (nucleus accumbens, ventromedial PFC) within ~140 ms; puppies and kittens actually out-rate human babies on cuteness across cultures because they over-satisfy the baby-schema template; fractal patterns with dimension D = 1.3-;1.5 produce roughly 60% stress reduction; and colour effects are quantifiable -; Valdez & Mehrabian even give a regression where brightness drives pleasure and saturation drives arousal.
There's a clear biophilic and cross-cultural thread: universal neural mechanisms set the baseline, and culture modifies them within milliseconds. That tension -; universal floor, cultural overlay -; is what made it interesting rather than just a list of "nice pictures".
How It Works
The scorer is a weighted ensemble of five analyzers, each one a direct port of a research finding into a measurable feature:
- Baby Schema (30%) -; Lorenz's Kindchenschema: face/eye detection, juvenile-feature proportions.
- Nature & Fractal (25%) -; biophilia plus fractal-dimension analysis, augmented with a Florence-2 vision model for scene understanding.
- Color Psychology (20%) -; the Valdez-;Mehrabian formulas mapping brightness/saturation to pleasure and arousal.
- Geometric Comfort (15%) -; curves-over-angles preference and symmetry.
- Facial Positivity (10%) -; Duchenne-smile detection.
On top of the score it generates per-analyzer visual overlays (eye-detection, green/blue nature highlighting, emotional color maps, curve-vs-angle maps, smile detection) and a batch mode that scans a folder, exports a CSV of metrics, and tracks processing performance.
Where It Landed
Archived as a working tool plus its research backbone. The scorer runs and produces interpretable breakdowns, which was the real goal -; not "is this image good" but "here's which evolved response it's pushing on, and how hard".
- Five-analyzer ensemble with weighted scoring and per-feature visualizations.
- Florence-2 integration for richer nature/scene detection.
- Batch processing with CSV export for running over a whole image library.