One of the fundamental problems in computer vision is the identification of objects from their shape. The research reported in this thesis is directed toward the development of a scheme for representing the shape of an object which allows it to be recognised both quickly and robustly across a wide range of viewing conditions. Given a shape described by a set of primitive elements, eg. straight line segments, the proposed scheme involves using a histogram to record the distribution of geometric features, eg. angle and distance, measured between pairs of primitives. This form of shape representation has a number advantages over previously proposed schemes. Foremost among these is the fact that it is able to produce local representations of shape, based on individual line segments. Recognition based on such representation is robust to the problems arising in cluttered scenes. Representations produced by the scheme are also invariant to certain object transformations, they degrade gracefully as the shape is fragmented and are strong enough to support discrimination between dissimilar objects. By treating the histogram recording a geometric feature distribution as a feature vector it is possible to match shapes using techniques from statistical pattern classification. This has the advantage that optimal matching accuracy can be achieved using processing which is both simple and uniform. The approach is therefore ideally suited to implementation in dedicated hardware. A detailed analysis is undertaken of the effect on recognition of changes in the description of a shape caused by fragmentation noise, scene clutter and sensor error. It is found that the properties of both the representation and matching components of the system combine to ensure that recognition is, in theory, unaffected by fragmentation noise, while it is maintained to very high levels of scene clutter. The factors which determine the effect of sensor error on the performance of the recognition system are fully analysed. The ability of the representational scheme to support object recognition is demonstrated in a number of different domains. The recognition of both 2D and 3D objects from a fixed viewpoint is demonstrated in conditions of severe fragmentation noise, occlusion and clutter. The scheme is then shown to extend straightforwardly to the representation of 3D shape. This is exploited to perform recognition and localisation of 3D objects from an arbitrary viewpoint, based on the matching of 3D scene and ,model shape descriptions. Finally, the use of the scheme within a multiple view-based approach to 3D object recognition is demonstrated.

No description available.

Information obtained from ultrasound images of fetal heads is often used to screen for various types of physical abnormality. In particular, at around 16 to 23 weeks' gestation two-dimensional cross-sections are examined to assess whether a fetus is affected by Neural Tube Defects, a class of disorders that includes Spina Bifida. Unfortunately, ultrasound images are of relatively poor quality and considerable expertise is required to extract meaningful information from them. Developing an ultrasound image recognition method that does not rely upon an experienced sonographer is of interest. In the course of this work we review standard statistical image analysis techniques, and explain why they are not appropriate for the ultrasound image data that we have. A new iterative method for edge detection based on a kernel function is developed and discussed. We then consider ways of improving existing techniques that have been applied to ultrasound Images. Storvik (1994)'s algorithm is based on the minimisation of a certain energy function by simulated annealing. We apply a cascade type blocking method to speed up this minimisation and to improve the performance of the algorithm when the noise level is high. Kass, Witkin and Terzopoulos (1988)'s method is based on an active contour or 'snake' which is deformed in such a way as to minimise a certain energy function. We suggest modifications to this energy function and use simulated annealing plus iterated conditional modes to perform the associated minimisation. We demonstrate the effectiveness of the new edge detection method, and of the improvements to the existing techniques by means of simulation studies.

Neural network analysis is proposed and evaluated as a method of analysis of marine biological data, specifically images of plankton specimens. The quantification of the various plankton species is of great scientific importance, from modelling global climatic change to predicting the economic effects of toxic red tides. A preliminary evaluation of the neural network technique is made by the development of a back-propagation system that successfully learns to distinguish between two co-occurring morphologically similar species from the North Atlantic Ocean, namely Ceratium arcticum and C. longipes. Various techniques are developed to handle the indeterminately labelled source data, pre-process the images and successfully train the networks. An analysis of the network solutions is made, and some consideration given to how the system might be extended.

Two Landsat TM scenes of SW England and a sub-scene of North Cornwall have been analysed visually in order to examine the effect of resolution on lineament interpretation. Images were viewed at several different scales as a result of varying image resolution whilst maintaining a fixed screen pixel size. Lineament analysis at each scale utilised GIS techniques and involved several stages: initial lineament identification and digitisation; removal of lineaments related to anthropogenic features to produce cleansed lineament maps; compilation of lineament attributes using ARC/INFO; cluster analysis for identification of lineament directional families; and line sampling of lineament maps in order to determine spacing. SW England lies within the temperate zone of Europe and the extensive agricultural cover and infrastructure conceal the underlying geology. The consequences of this for lineament analysis were examined using sub-images of North Cornwall. Here anthropogenic features are visible at all resolutions between 30m and 120m pixel sizes but lie outside the observation threshold at 150m. Having confidence that lineaments at this resolution are of non-anthropogenic origin optimises lineament identification since the image may be viewed in greater detail. On this basis, lineament analysis of SW England was performed using image resolutions of 150m. Valuable geological information below the observation threshold in 150m resolution images is likely, however, to be contained in the lineament maps produced from higher resolution images. For images analysed at higher resolutions, therefore, knowledge-based rules were established in order to cleanse the lineament populations. Compiled lineament maps were 'ground truthed' (primarily involving comparison with published geological maps but included phases of field mapping) in order to characterise their geological affinities. The major lineament trends were correlated to lithotectonic boundaries, and cross-cutting fractures sets. Major lineament trends produced distinct frequency/orientation maxima. Multiple minor geological structures, however, produced semi-overlapping groups. A clustering technique was devised to resolve overlapping groups into lineament directional families. The newly defined lineament directional families were further analysed in two ways: (i) Analysis of the spatial density of the length and frequency of lineaments indicates that individual and multiple lineament directional families vary spatially and are compartmentalised into local tectonic domains, often bounded by major lineaments. Hence, such density maps provide useful additional information about the structural framework of SW England. (ii) Lineament spacing and length of the lineament directional families were analysed for the effect of scale and geological causes on their frequency/size distributions. Spacing of fracture lineaments were found to be power-law, whereas lengths showed power-law and non-power-law distributions. Furthermore the type of frequency/size distribution for a lineament directional family can change with increasing resolution.

No description available.

No description available.

No description available.

No description available.

No description available.