Thesis (Ph. D.)--University of Wisconsin--Madison, 1961. / Typescript. Vita. eContent provider-neutral record in process. Description based on print version record. Includes bibliographical references.

When computer generated images are used for real-time display in simulator applications, much of the fine detail available from the natural world, or even from video-film, is not available to an observer. This lack of detail leads to a reduction in the number of sources of depth information (cues to depth) that are available to specify the layout of the displayed scene. Amongst the cues normally available are luminance gradients and luminance contrast gradients, each deriving from luminance differences between components of the displayed scene; however, in computer generated images, these two cues do not always conform to the intended natural world image, and can offer conflicting information. While not referring explicitly to luminance gradients, Ames (1949) demonstrated that the brighter of two otherwise identical objects would appear nearer; his Demonstration 18 offering a negative luminance gradient similar to that arising in the natural world from atmospheric perspective. Similarly, Ross (1967, 1993) and O'Shea, Blackburn and Ono (1994) have shown a similar effect to Ames (1949), but with higher contrast replacing increased brightness, which they liken to the negative luminance contrast gradients that are also available in the natural world due to atmospheric perspective. The luminance gradient, and luminance contrast gradient cues are generally in accord when the scene background is light, but are in conflict where the background is dark. The experiments reported here show that either gradient can function as a cue to depth, and hence to the spatial layout of a depicted scene, and that conflicts between them are resolved in a way that takes into account the amount and type of other depth information available to an observer. Such a form of conflict resolution and cue combination is in accord with the separate items of depth information being processed either partly or wholly in parallel, so that the strength of each cue is determined by reference to the other available cues. When applied to simulators using computer generated images, these results suggest that both users, and scenario designers, require an awareness of the possible effect of a change to any item of depth information, and in particular to depth information that has its origin in luminance differences between objects in the depicted scene.

Colour constancy is defined as the ability to perceive the surface colours of objects within scenes as approximately constant through changes in scene illumination. Colour constancy in real life functions so seamlessly that most people do not realise that the colour of the light emanating from an object can change markedly throughout the day. Constancy measurements made in simple scenes constructed from flat coloured patches do not produce constancy of this high degree. The question that must be asked is: what are the features of everyday scenes that improve constancy? A novel technique is presented for testing colour constancy. Results are presented showing measurements of constancy in simple and complex scenes. More specifically, matching experiments are performed for patches against uniform and multi-patch backgrounds, the latter of which provide colour contrast. Objects created by the addition of shape and 3-D shading information are also matched against backgrounds consisting of matte reflecting patches. In the final set of experiments observers match detailed depictions of objects - rich in chromatic contrast, shading, mutual illumination and other real life features - within depictions of real life scenes. The results show similar performance across the conditions that contain chromatic contrast, although some uncertainty still remains as to whether the results are indicative of human colour constancy performance or to sensory match capabilities. An interesting division exists between patch matches performed against uniform and multi-patch backgrounds that is manifested as a shift in CIE xy space. A simple model of early chromatic processes is proposed and examined in the context of the results.

Diss.--Philosophische Fakultät II--Würzburg--Bayerische Justus-Maximilians-Universität, 1997. / Bibliogr. p. 347-394. Index.

Issued also as Thesis (Ph. D.)--Columbia University. / Bibliography: p. 86-88.

Th. 3e cycle--Sci. du comportement--Aix-Marseille 2, 1977.

The desire to produce artificial vision systems which behave in an intelligent, humanlike way or which can autonomously and automatically perform tasks currently only performed by humans has been a goal of Artificial Intelligence research for many decades. Until recently much of the research concentrated on extracting visual representations of objects from single, static scenes. The last decade has seen an increase in interest concerning mobile robotics for navigation, planning and autonomous control as well as for the interpretation of events in real, dynamic scenes. Presented in this thesis is research on artificial vision systems from two different, but both necessary, standpoints. One concerns low-level vision-based behaviour of object tracking based upon a naturalistic theory of perception and behaviour within living systems. The other takes a more application and engineering based approach and its goal is to address high-level scene interpretation and control of processing resources. Numerous experiments are presented to demonstrate the various issues. The two main experiments, corresponding to the two research streams, are a system which is able to fixate complex multi-coloured objects and a fully integrated vision system for predicting and following, with a mobile sensor, events in a dynamic scene.

A common problem in vision research is explaining how humans perceive a coherent, detailed and stable world despite the fact that the eyes make constant, jumpy movements and the fact that only a small part of the visual field can be resolved in detail at any one time. This is essentially a problem of integration over time - how successive views of the visual world can be used to create the impression of a continuous and stable environment. A common way of studying this problem is to use complete visual scenes as stimuli and present a changed scene after a disruption such as an eye movement or a blank screen. It is found in these studies that observers have great difficulty detecting changes made during a disruption, even though these changes are immediately and easily detectable when the disruption is removed. These results have highlighted the importance of motion cues in tracking changes to the environment, but also reveal the limited nature of the internal representation. Change blindness studies are interesting as demonstrations but can be difficult to interpret as they are usually applied to complex, naturalistic scenes. More traditional studies of scene analysis, such as visual search, are more abstract in their formulation, but offer more controlled stimulus conditions. In a typical visual search task, observers are presented with an array of objects against a uniform background and are required to report on the presence or absence of a target object that is differentiable from the other objects in some way. More recently, scene analysis has been investigated by combining change blindness and visual search in the `visual search for change' paradigm, in which observers must search for a target object defined by a change over two presentations of the set of objects. The experiments of this thesis investigate change blindness using the visual search for change paradigm, but also use principles of design from psychophysical experiments, dealing with detection and discrimination of basic visual qualities such as colour, speed, size, orientation and spatial frequency. This allows the experiments to precisely examine the role of these different features in the change blindness process. More specifically, the experiments are designed to look at the capacity of visual short-term memory for different visual features, by examining the retention of this information across the temporal gaps in the change blindness experiments. The nature and fidelity of representations in visual short-term memory is also investigated by manipulating (i) the manner in which featural information is distributed across space and objects, (ii) the time for which the information is available, (iii) the manner in which observers must respond to that information. Results point to a model in which humans analyse objects in a scene at the level of features/attributes rather than at a pictorial/object level. Results also point to the fact that the working representations which humans retain during visual exploration are similarly feature- rather than object-based. In conclusion the thesis proposes a model of scene analysis in which attention and vSTM capacity limits are used to explain the results from a more information theoretic standpoint.

Making a robot autonomous has been a common challenge to be overcome since the very beginning. To be an autonomous system, the robot should collect environmental data, interpret them, and act accordingly. In order to accomplish these, some resource management should be conducted. That is, the resources, which are time, and computation power in our case, should be allocated to more important areas. Existing researches and approaches, however, are not always human like. Indeed they don&rsquo / t give enough importance on this. Starting from this point of view, the system proposed in this thesis supplies the resource management trying to be more &rsquo / human like&rsquo / . It directs the focus of attention to where higher resolution algorithms are really needed. This &rsquo / real need&rsquo / is determined by the visual features of the scene, and current importance levels (or weight values) of each of these features. As a further attempt, the proposed system is compared with human subjects&rsquo / characteristics. With unbiased subjects, a set of parameters which resembles a normal human is obtained. Then, in order to see the effect of the guidance, the subjects are asked to concentrate on a single predetermined feature. Finally, an artificial neural network based learning mechanism is added to learn to mimic a single human or a group of humans. The system can be used as a preattentive stage module, or some more feature channels can be introduced for better performance in the future.

Thesis (Ph. D.)--University of Wisconsin--Madison, 1968. / Typescript. Vita. Includes bibliographical references.