| In this thesis different aspects of human visual processing were investigated with respect to search and inspection for machine vision applications. Preliminary data were taken on dynamic search and inspection tasks where it was observed that a significant amount of the time (7%) the observer was able to locate a target beyond a 5{dollar}spcirc{dollar}-eccentricity boundary where pop-out effects normally do not occur. Using the empirical data from these results, a probabilistic model of the dynamic search task was derived, which indicated that the observer's performance for peripheral targets was respectably above chance. Thus, these empirical and statistical findings incited the examination of the role of attention and visual processing in search and inspection. Extending on a Saarinen and Julesz paradigm where they examined the speed of attentional shifts in the visual field, in addition to a Shiffrin et al. paradigm where they studied aspects of attentional control and information capacity limitations, a two-phase study was designed to investigate further aspects of search and inspection at the cognitive level. In Phase I, the processing of numerals presented sequentially and simultaneously was studied. In Phase II, more realistic stimuli, such as fruit, musical instruments, and tools, were used to study the effect of context on processing, in addition to extending on the Phase I findings. By using common everyday objects, the results would be more applicable to natural search and inspection situations encountered by humans and machines. The major findings indicated that at the earliest stages of visual processing, attention might not be a major factor and the "information bottleneck" in processing was most likely due to the data-driven mechanisms that control the input into short-term memory. Also, in Phase II of the study, empirical evidence indicated that top-down semantic processes most likely interact with data-driven bottom-up processes to increase the efficiency of the"information bottleneck." This was significant because, though theoretical models of visual processing commonly assumed that there is interaction between top-down and bottom-up processes, to date very little empirical evidence has been obtained to prove or disprove this position. |
