Meaning and uncertainty inherent in understanding images, spatial-taxon hierarchy, word annotation and relevant context

Abstract

This paper explores the meaning and uncertainty inherent in (a) understanding image hierarchies; (b) describing them with words; and (c) navigating the abstraction context of the viewer. A spatial-taxon hierarchy, a standardized scene architecture, partitions an image into a foreground, subject and salient objects and/or sub-objects. The introduction starts with a thought experiment (Thought experiments, borrowed from the model-theoretic isomorphism standard of structure-mapping theory, enable readers to compare two systems thought to be similar. It’s a form of inductive reasoning that expands knowledge in the face of uncertainty (Holland et al. 1986 [13]) by providing an explicit representation of how two systems are similar. Though the conclusion that the two systems do share an isomorphic structure can only be supported via various degrees of truth (fuzzy membership), it establishes its plausibility. Analogical reasoning is natural to human thought and communication making it useful for scientific papers.) based on a poem & an image landscape. The thought experiment is intended to provide analogical inference as scaffolding for the rest of the paper. The results of experimental data of human annotated spatial-taxon and corresponding word descriptions of two images are presented. The experimental results are analyzed in terms of spatial-taxon designation and the meaning & uncertainty presented by the human annotations. The results support the fuzzy spatial-taxon hierarchy of human scene perception described by other works, show that word descriptions depend on spatial-taxon designation and that long tail word distributions require unbounded possibility with semantic uncertainty (type 2 fuzzy sets) for the word counts in the probability distribution. Deep learning image recognition, Zadeh information restriction principal, Shannon’s distinction between information content and semantics, customized image descriptions and fuzzy inference techniques are explored.