Image Object Analysis
Object Detection API
Both raw accuracy and interpretive value were assessed, including precision, recall, and the relationship between model confidence and correctness. Google Vision was more precise and conservative, excelling in well-defined domains but often missing symbolically relevant or ambiguous content. Clarifai captured a broader, more nuanced range of objects— ncluding key ecological and symbolic features, though it showed a higher rate of overconfident false positives.
Statistical threshold analysis determined that a confidence score of 0.87 optimally balanced precision (91%) and recall (95.5%) for Clarifai, enabling robust inclusion of relevant labels while minimising false positives. Clarifai was therefore selected as the primary model for subsequent analysis, with all outputs filtered by the calibrated threshold to ensure semantic richness and methodological reliability.
Object Grouping
Detected objects from 7,362 images were classified independently into two separate systems: ten Developmental Affordance (DA) object groups and six Maslow Hierarchy of Needs (MHN) levels. Each object was assigned to a single group within either framework DA or MHN using only concrete, visible criteria, and the two classification processes were entirely distinct.
|
DA Object Groups:
|
Maslow Hierarchy of Needs (MHN) Groups:
|
Critical Integration and Limitations of Object Grouping
Out of 116,944 objects detected, only 43,579 (37.3%) could be successfully classified into either system, with 73,365 objects (62.7%) remaining uncategorised due to factors such as abstraction, ambiguity, or low detection confidence. Only 2,239 objects (1.9%) were classified by both systems, underscoring the conceptual divergence between Maslow and DA groupings.
The majority of classified objects were unique to one framework (Maslow-only: 20,121; DA-only: 21,219), reflecting the complementary yet limited scope of each taxonomy. These findings illustrate both the analytical potential and the inherent constraints of applying structured psychological models to complex, culturally varied image data, and underscore the need for cautious interpretation of results.