Image Analysis
Rationale for Image Analysis
This project uses large-scale image scraping to explore how fundamental human needs such as play, nature, safety, and creativity are visually expressed in digital culture. Images act as a parallel language, conveying emotional and instinctive priorities that written policy often misses (Barthes, 1977; Barry, 1997; Rose, 2022).
By analysing which images surface for developmental keywords, the research examines whether digital visual culture aligns with, or diverges from, our evolutionary needs for connection and wellbeing (Bjorklund & Pellegrini, 2000; Haidt, 2024; Michaelson et al., 2020). Everyday digital photography becomes a kind of collective intuition about what matters, revealing unmet needs in contemporary life (Hand, 2012; Maslow, 1943).
Image Scrape
While this strategy enhanced the thematic relevance of the images, it also introduced methodological bias by excluding certain symbolic or artistic representations. This approach was guided by the project’s aim to prioritise embodied, ecological, and evolutionary aspects of human thriving. Approximately 300–500 images were collected per term, and stored for further analysis using visual recognition tools.
This methodology is not neutral, and future research could compare filtered and unfiltered results to reveal both dominant and marginal visual narratives. All image collection was for educational, non-commercial use, under UK fair dealing.
Image Object Analysis
Automated object recognition was evaluated using both Google Vision API and Clarifai, with Clarifai selected for its broader semantic coverage following calibration at a confidence threshold of 0.87 (yielding 91% precision and 95.5% recall). Detected objects from 7,362 images were classified independently into ten Developmental Affordance (DA) categories or six Maslow Hierarchy of Needs (MHN) levels, using distinct, concrete criteria.
Of 116,644 objects detected, only 37.3% could be grouped within either system; the majority were unique to a single framework, with just 1.9% overlapping both. These results underscore the complementary strengths and limitations of each taxonomy, and the challenges of structured classification in complex image datasets.