Use this URL to cite or link to this record in EThOS:
Title: Detection of dynamic form in faces and fire
Author: Nagle, F. S.
ISNI:       0000 0004 8503 9616
Awarding Body: UCL (University College London)
Current Institution: University College London (University of London)
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
Moving natural scenes pose a challenge to the human visual system, containing diverse objects, clutter, and backgrounds. Well-known models of object recognition do not fully explain natural scene perception, ignoring segmentation or the recognition of dynamic objects. In this thesis, we use a familiar natural stimulus, moving flames, to evaluate the human visual system's ability to match and search for complex examples of dynamic form. What can analysis in the image domain tell us about dynamic flame? Using image statistics, Fourier analysis and motion evaluation algorithms, we analysed a highresolution dataset typical of moving flame. We characterise it as a motion-rich stimulus with an exponential power spectrum and few long-range spatial or temporal correlations. Are observers able to effectively encode and recognise dynamic flame stimuli? What visual features play an important role in matching? To investigate, we set observers matching tasks using clips from the same dataset. Colour changes do not affect matching on short clips, but inversion and reversal do. We show that dynamic edges are a key component of flame representations. Can observers search well for flame stimuli? Can they detect targets (short flame clips) in equally-sized longer clips? Using temporal search tasks, we show that observers' accuracy drops quickly as the search space grows; there is no pop-out. Accuracy is not so strongly affected by a blank ISI, however, showing that search difficulties, rather than representational decay, are to blame. In conclusion, we find that the human visual system is capable of matching the complex motion patterns of dynamic flame, but finds search much harder. We find no evidence of category orientation specialisation. Combining several experimental results, we suggest that the representation of dynamic flame is neither snapshot-based nor dedicated and high-level, but relies on the encoding of sparse, local spatiotemporal features.
Supervisor: Johnston, A. ; McOwan, P. W. Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available