Visual Modeling by Contextual Influences in Environmental Information Systems

Global IP Fellows Meeting Home Agenda Participants Presentations Star Hill Weekend Star Logo links Event Photos

Nelley Kovalevskaya
Senior Researcher, Institute for Water and Environmental Problems, Barnaul, Russia
Visual Modeling by Contextual Influences in Environmental Information Systems

Visual Modeling by Contextual Influences in Environmental Information Systems

The vision-knowledge-based interpretation of environment states by remotely sensed imagery is still more art than a formal theory. It is mostly descriptive, uses fuzzy terms and is not systematically equated with the measurable attributes. The purpose of the study is to gain more insight into the processes of knowledge capture connected with changes in environmental patterns and reformulate this knowledge into quantitative terms. Analysis of spatial attributes appears to be necessary in most areas where the differences in spatial data between regions in the image occur. A single image region, or a pattern, is assumed to be defined by the homogeneity or translation invariance of the statistics of the image features. A key question of knowledge sharing is what an environment expert perceives from the pattern. More precisely, what are the sufficient and necessary features so that a pair of patterns sharing such features can be regarded as the ”same class” for different experts. Assume that these conspicuous (highlighting) visual features are perceived to be a sketch of the pattern known for an expert. Such environmental perceptual sketch (EPS) would have obvious significance for visual knowledge sharing. On computational level the problem of highlighting features’ representation means modeling of the pattern homogeneity. Let us define expert perception of homogeneous pattern of environment as popout of a set of targets, which are situated on the pattern as much regularly as the pattern is homogeneous. In this case an important issue of EPS-modeling is opportunity to be used by natural intelligences in constructing of images and patterns of environment states due to “biological plausibility”. The last one means that EPS-model is consistent with previous experimental evidences from another disciplines (Mimford, 1992; Treisman et al, 1980; Scalfia and Joffe, 1995). It is natural to assume that an expert perceives the most effectively those signals that occur most frequently. Thus, it is statistical properties of the environment that are relevant for sensory processing. On the other hand, environmental patterns contain characteristic statistical regularities that set them apart from purely random patterns, or independent random fields (IRF). Let us suppose that the more regular form of structure containing in environment pattern, the more effective sketch of the pattern for higher level operations and coding into memory of an expert. There should be a parameter of the EPS-modeling to measure regularity of pattern structure, or distance from IRF-pattern. For example, it’s obvious that Pattern1 is farther from IRF-pattern than Patter2 (Fig.2). In case of EPS-modeling pattern regularity measure, or non-resemblance with IRF, ranges all natural patterns through validation of expert perception comparisons. The success of EPS-approach depends on three choice-issues: A. Choice on model of PSP that enable to adjust the pre-attentive and attentive visual systems on computational level and to represent continuum of visual mechanisms depending on local (long- and short-range) interactions can contribute to global view. B. Choice on the most rich theoretical framework, in which the model (A) above could be embedded. C. Choice on a function (functions) of measure of pattern regularity, or distance from IRF-pattern, which is concerted with model (A) and theory of optimal decision (B) as well as expert perception of homogeneous pattern.

D.

                        IRF-pattern               Pattern1                  Pattern2

                          0                                                        Measure of pattern regularity (non-resemblance with IRF)
 

Experiments were carried out with the images of Siberia (Russia). Many natural patterns appear to be modeled adequately by the proposed model. The validity of this model can be visually and quantitatively checked by comparing simulated samples with the training one. The spatial homogeneity or piecewise homogeneity of a training sample can also be quantitatively verified by matching sample relative frequency distributions of gray level combinations collected over different patches within the sample. Sharing environmental knowledge can benefit from EPS-approach because such kind of encoding reduces the data rate without significant information loss, moreover the level of loss may be managed by parameters of proposed model. So, significant saving can be made by avoiding transmitting the information redundantly. In this case there is no problem of how does one judge whether or not accept the fit of the environment pattern sketch and decide that the signal does contain a valid sample of the pattern in question. In fact, the sketch appears to be a concise representation of the environment region and creates more salient locations. So, the model is needed for characterizing the spatial arrangements of the components. The discovery of meaningful components could go away beyond pairwise interactions, including statistics Nth order. For example, triplets, quadruples of elements, etc could be used in the next stage. EPS -modeling leads to a better understanding of expert capacity for fast categorization of environmental objects. References 1. N.Kovalevskaya and V.Pavlov.2002. Environmental Mapping Based on Spatial Variability, in: Journal of Environmental Quality, 31:1462-1470 (2002). 2. Kovalevskaya N. 2002. Landscape Indication Based on Stochastic Relaxation. P.121-146. In Ranian S. Muttiah (Ed.). From Laboratory Spectroscopy to Remotely Sensed Spectra of Terrestrial Ecosystems, Kluwer Academic Publishers. 3. Mumford. 1992. On the computational architecture of the neocortex, II The role of cortico-cortical. loops, Biological Cybernetics, 66, 241-251. 4. Scialfa C T and Joffe. 1995. KM1995 Preferential processing of target features in texture segmentation Percept. Psychophys. 57 1201–8 5. Treisman and G. Gelade. 1980. A feature integration theory of attention. Cognitive Psychology, 12:97--136.