Peter Kovesi

Shapelets correlated with surface normals produce surfaces

This paper addresses the problem of deducing the surface shape of an object given just the surface normals. Many shape measurement algorithms such as shape from shading and shape from texture only return the surface normals of an object, often with an ambiguity of pi in the surface tilt. The surface shape has to be inferred from these normals, typically via some integration process. However; reconstruction through the integration of surface gradients is sensitive to noise and the choice of integration paths across the surface. In addition, existing techniques cannot accommodate ambiguities in tilt. This paper presents a new approach to the reconstruction of surfaces from surface normals using basis functions, referred to here as shapelets. The surface gradients of the shapelets are correlated with the gradients of the surface and the correlations summed to form the reconstruction. This results in a simple reconstruction process that is very robust to noise. Where there is an ambiguity of it in the surface tilt, reconstructions of reduced quality are still possible up to a positive/negative shape ambiguity. Intriguingly, some form of reconstruction is also possible using just slant information

Towards the automated analysis of regional aeromagnetic data to identify regions prospective for gold deposits

Quantitative analysis of geoscientific data to determine areas most likely to contain mineral deposits is becoming increasingly common in the mining industry. The approach is based on characterising areas known to contain deposits and seeking similar areas elsewhere. This paper presents an automatic image processing technique for the prospectivity analysis of Archaean lode-gold deposits, which differs from previous methods in that it is based solely on aeromagnetic data and does not require knowledge of the location of existing deposits. Instead, the aeromagnetic expressions of what are perceived to be geologically significant characteristics are sought within the aeromagnetic data. Gold mineralisation is known to occur near major crustal breaks manifesting as large-scale shear zones, which act as conduits for mineralising fluids. Mineralisation occurs in regions of structural complexity adjacent to the shear zones. Progressing towards the automatic detection of such regions, the proposed system finds firstly regions of magnetic discontinuity that correspond to both lithological boundaries and shear zones using a combination of texture analysis and symmetry feature detection techniques. Secondly, it examines the data using fractal analysis to find areas nearby with a complex magnetic expression (zones of structural complexity). The most prospective areas are those where inferred structural complexity occurs adjacent to the regions of magnetic discontinuity. A preliminary experiment was conducted using aeromagnetic data from the Yilgarn Craton in Western Australia and the regions selected by the proposed system contained over 76% of all known deposit locations and 82% of the greater than 1 tonne deposit locations.

Fast Almost-Gaussian Filtering

Image averaging can be performed very efficiently using either separable moving average filters or by using summed area tables, also known as integral images. Both these methods allow averaging to be performed at a small fixed cost per pixel, independent of the averaging filter size. Repeated filtering with averaging filters can be used to approximate Gaussian filtering. Thus a good approximation to Gaussian filtering can be achieved at a fixed cost per pixel independent of filter size. This paper describes how to determine the averaging filters that one needs to approximate a Gaussian with a specified standard deviation. The design of bandpass filters from the difference of Gaussians is also analysed. It is shown that difference of Gaussian bandpass filters share some of the attributes of log-Gabor filters in that they have a relatively symmetric transfer function when viewed on a logarithmic frequency scale and can be constructed with large bandwidths.

Phase Preserving Tone Mapping of Non-Photographic High Dynamic Range Images

Non-photographic images having a high dynamic range, such as aeromagnetic images, are difficult to present in a manner that facilitates interpretation. Standard photographic high dynamic range (HDR) algorithms may be unsuitable, or inapplicable to such data. We present a method that compresses the dynamic range of an image while preserving local features. It makes no assumptions about the formation of the image, the feature types it contains, or its range of values. Thus, unlike algorithms designed for photographic images, this algorithm can be applied to a wide range of scientific images. The method is based on extracting local phase and amplitude values across the image using monogenic filters. The dynamic range of the image can then be reduced by applying a range reducing function to the amplitude values, for example taking the logarithm, and then reconstructing the image using the original phase values. An important attribute of this approach is that the local phase information is preserved, this is important for the human visual system in interpreting the image. The result is an image that retains the fidelity of its features within a greatly reduced dynamic range. An additional advantage of the method is that the range of spatial frequencies that are used to reconstruct the image can be chosen via high-pass filtering to control the scale of analysis.

MATLAB as an introductory programming language

This paper explores the use of MATLAB for teaching the fundamental constructs of programming languages to engineering and science students. It was found that the time taken for students to learn programming can be reduced by up to 50%, paving the way for using the language to solve problems.

Interactive multi-image blending for visualization and interpretation

The need to integrate information from images of different modalities is an increasingly common problem for the geosciences. Interactive multi-image blending is presented as a tool for facilitating the interpretation of complex information from multiple data sources. Traditionally, image blending has only been considered for cross-dissolving effects between two images. The emphasis of this work is on image blending for the effective visualization of data, rather than for attractive visual effects. To achieve this we have developed blending techniques that allow for the simultaneous presentation of more than two images. We present a family of different image blending techniques that support the blending of multiple images under a range of different situations. For image blending to be a useful tool for data interpretation it is important that the input images remain distinct within the blend. We argue that interactivity of the blend is an important component for achieving this. Blending can also be usefully employed to interactively explore parameter variations for enhancement techniques. Often the best parameter values to use cannot be known beforehand, and it is common for different regions of an image to require different parameter values for best enhancement. HighlightsGeological interpretation requires integration of data from multiple sources.Interactive blending of two or more images facilitates interpretation.We present a series of image blending techniques designed for different applications.Interactivity is crucial for ensuring input images remain identifiable within a blend.Blending can be used to interactively explore parameter variations for enhancement.

Trajectory Based Video Sequence Synchronization

Video sequence synchronization is often necessary for computer vision applications where multiple simultaneously recorded videos are processed. We present a coarse-to-fine approach to synchronizing two video sequences recorded at the same frame rate by stationary cameras with fixed internal parameters. At the coarse level, each sequence is broken into a set of sub-sequences, which are then matched. A voting scheme determines the range in which the sequences’ temporal offset lies. The fine synchronization step searches for the temporal offset by initially examining integer offsets, and then using the golden-section search to locate the offset to sub-frame accuracy. Our algorithm recovers the temporal offset of two sequences using the motion of a single moving object, is computationally efficient, and it does not require any stationary background points as reference points. We present results for synthetic data and real video sequences, with various degrees of temporal overlap.

Projective Transformations for Image Transition Animations

Transformation of image patches is a common requirement for 2D transition animations such as shape interpolation and image morphing. It is usually done by applying affine transformations to triangular patches. However, the affine transformation does not model the perspective transformation frequently found in images. Hence, such techniques can only produce approximate results and usually use an excessively large number of triangles to compensate for this shortcoming. This paper proposes the application of projective transformations on quadrilateral image patches as a solution to this problem. We address the issues of appropriate decomposition and interpolation of projective transformation matrices to produce a natural looking transition animation for a single quadrilateral as well as for shapes made up of multiple quadrilaterals.

A 3D local energy surface detector for confocal microscope images

The ability to detect features within confocal microscope images is important for the interpretation and analysis of such data. Most detectors are gradient based, and so are sensitive to noise, and fail to accurately locate some feature types that are important in confocal microscopy. The local energy feature detector developed by M.C. Morrone and R.A. Omens (1987) marks locations where there is maximal congruence of phase in the Fourier components of an image. Points of maximal phase congruency occur at all common feature profiles: step and roof edges, line features and Mach bands. A 3D implementation of the local energy feature detector, suitable for confocal microscope data, is presented. The detector computes local energy by convolving an image with oriented pairs of 3D filters. The filters are 3D versions of Morlet wavelets. To increase the speed of the convolution, the filters are designed in frequency space and multiplied by the images Fourier transform. Results are presented for real confocal images and synthetic 3D image volumes.

Motion guided video sequence synchronization

We present an algorithm that synchronizes two short video sequences where an object undergoes ballistic motion against stationary scene points. The object’s motion and epipolar geometry are exploited to guide the algorithm to the correct synchronization in an iterative manner. Our algorithm accurately synchronizes videos recorded at different frame rates, and takes few iterations to converge to sub-frame accuracy. We use synthetic data to analyze our algorithm’s accuracy under the influence of noise. We demonstrate that it accurately synchronizes real video sequences, and evaluate its performance against manual synchronization.