This paper deals with the order in which different levels of form are recognized in a visual image. An experiment is reported in which the size of a tachistoscopically viewed image was varied. The results suggest neither an invariant "top-down" (gross shapes first followed by lower-order details) or "bottom-up" (the opposite) sequence. Rather, they seem to suggest a sort of "middle-out" sequence: forms at some intermediate level of structure having an optimal size or spatial-frequency spectrum are processed first, with subsequent processing of both higher and lower levels of form.Contemporary theories of visual information processing have been strongly influenced by work on computer pattern recognition, particularly the concept of image structure employed in syntactic scene analysis (see Fu, 1974). The idea is that an image (a "scene") can be parsed into hierarchical levels of form, much as a paragraph can be parsed into sentences, phrases, words, etc. The value of such a representation stems from the redundancy (correlation) between different levels of structure in most scenes. For example, the fact that a scene contains a "head" is highly correlated with the presence of such lowerorder components as "eyes," "ears," and "nose," etc. Knowledge of such structural redundancy can facilitate computer processing of images, just as syntactical redundancy facilitates the processing of language.This redundancy can be utilized in a "top-down" sequence of processing, where the identification of a higher-order form (e.g., head) facilitates its subsequent analysis into lower-order components (e.g., eyes, ears, etc.) or in a "bottom-up" sequence, whereby identification of the lower-order components facilitates their subsequent synthesis into the higher-order form. Redundancy has also been employed in more complicated sequences involving both top-down and bottomup components, for example, "analysis by synthesis," in which a tentative synthesis of components into a higher-order form facilitates subsequent analysis of previously ambiguous components (e.g., Halle & Stevens, 1962).

No abstract

Two experiments are reported in which observers had to utilize information from one of two structural levels of visual stimulus patterns (large letters composed of smaller ones). They could utilize information more rapidly form one level only at the cost of slower utilization from the other. This tradeoff defines an empirical attention operating characteristic (AOC) which is consistent with a simple mathematical model of the perceptual process: when viewing a stimulus, the observer selects one of two alternative "attentional" strategies, where each strategy is optimal for utilizing information from one structural level, but less than optimal for the other.

A model is presented of the perceptual process through which an observer compares two consecutively observed stimuli. Emphasis is placed on the manner in which a memory of the first stimulus is maintained until the comparison stimulus is observed. It is argued that the role of this perceptual memory process provides the pr.imary distinction between detection and recognition tasks. Two experiments are reported: an experiment in which the observer is asked to judge the similarity in position of two points of light presented serially in a dark room; and an experiment in which the observer judges the similarity in loudness of two 'serially presented tones. The visual experiment is discussed in relation to theanalysis of autokinesis and involuntary eye movements, while the auditory experiment is shown to have special relevanc e to the issue of time-order err ors.To distinguish a difference between two stimuli an observer must have a receptor or input process which will respond differently to the two stimuli. Furthermore, to compare two stimuli presented at different points in time, he must maintain a memory of the first stimulus until the second one occurs. For example, suppose an observer were asked to judge the similarity in loudness of two serially presented tones. A minimal requirement for accurate discrimination would be a receptor process which reacted differently to the two tone amplitudes. Furthermore, if the second tone occurred with some delay following cessation of the first tone, the observer would have to maintain a memory of the initial tone's loudness until he heard the second tone. Thus one might say that discrimination of serially observed stimuli is limited by at least two factors: one, the resolving power of the receptor or input process; and two, the efficiency of the memory process through which the observer retains information about one stimulus until the comparison stimulus occurs.Recent theoretical work on the psychophysics of detection has emphasized the role of the input process in determining sensory limits. The most prominent of these models is the psychophysical Theory of Signal Detection (see Green & Swets, 1966, for a comprehensive development of this theory). This is a twoprocess theory in that detection is represented as the product of two subprocesses, an input process and a decision or response process. The input proce ss specifies how the stimulus event evokes one of two sets of hypothetical sensory states in the observer; the decision process specifies how a given sensory state determines the observer's overt response. The particular sensory state evoked on each trial is treated as a value of a random variable whose distribution depends on the stimulus value. If two stimulus values evoke similar distributions of sensory states, it is difficult for the observer to distinguish which stimulus has given rise to a particular sensory state. His decision is assumed to be a statistical inference about the stimulus value based on his "sensory sample," the sensory state, and his knowledge o...

No abstract