David Pycock

A Scale-Space Medialness Transform Based on Boundary Concordance Voting

The Concordance-based Medial Axis Transform (CMAT) presented in this paper is a multiscale medial axis (MMA) algorithm that computes the medial response from grey-level boundary measures. This non-linear operator responds only to symmetric structures, overcoming the limitations of linear medial operators which create “side-lobe” responses for symmetric structures and respond to edge structures. In addition, the spatial localisation of the medial axis and the identification of object width is improved in the CMAT algorithm compared with linear algorithms. The robustness of linear medial operators to noise is preserved in our algorithm. The effectiveness of the CMAT is accredited to the concordance property described in this paper. We demonstrate the performance of this method with test figures used by other authors and medical images that are relatively complex in structure. In these complex images the benefit of the improved response of our non-linear operator is clearly visible.

Robust statistical models for cell image interpretation

A robust and adaptable model-based scheme for cell image interpretation is presented that can accommodate the wide natural variation in the appearance of cells. This is achieved using multiple models and an interpretation process that permits a smooth transition between models. Boundaries are represented using trainable statistical models that are invariant to transformations of scaling, shift, rotation and contrast; a Gaussian and a circular autoregressive (CAR) model are investigated. The interpretation process optimises the match between models and data using a Bayesian distance measure. We demonstrate how objects that vary in both shape and grey-level pattern can reliably be segmented. The results presented show that overall performance is comparable with that for manual segmentation; the area within the automatically and manually selected cell boundaries that is not common to both is less than 5% in 96% of the cases tested. The results also show that the computationally simpler Gaussian boundary model is at least as effective as the CAR model.

Robust model-based boundary cue generation for cell image interpretation

In model based image interpretation the cue generation process must be robust; i.e. able to generate appropriate cues for wide variety of objects. In this work we describe a model-based approach to cue generation and demonstrate its operation on both synthetic test data and natural images of epithelial cells. Epithelial cells were chosen because their appearance is highly variable. Statistical inference is used to interpret step, ramp and composite grey-level edges where an edge is modelled in terms of the greylevel distributions on each side of the edge. This approach provides a measure of confidence for each edge cue generated. The cue generation process reported here is shown to be more selective than other, previously reported methods that in turn compare favourable with the performance of the Canny operator for cue generation. The cue generation process reported here is also shown to be robust for free laying, touching and overlapping cells when combined with a simple boundary interpretation strategy.

CBIT - context-based image transmission

Few networks offer sufficient bandwidth for the transmission of high resolution two and three-dimensional medical image sets without incurring significant latency. Traditional compression methods achieve bit-rate reduction based on pixel statistics and ignore visual cues that are important in identifying visually informative regions. The paper describes an approach to managing image transmission in which spatial regions are selected and prioritized for transmission so that visually informative data is received in a timely manner. This context-based image transmission (CBIT) scheme is a lossless form of progressive image transmission (PIT) in which gross structure, represented by an approximate iconic image, is transmitted first. Each part of this iconic image is progressively updated, using a simple set of rules that take into account viewing requirements. CBIT is realized using knowledge about image composition to segment, label, prioritize, and fit geometric models to regions of an image. Tests, using neurological images, show that, with CBIT, a valuable transmitted image is received with a latency that is about one-tenth that of traditional PIT schemes. Frequently, the necessary regions of the image are transmitted in about half the time taken to transmit the full image.

Robust model-based signal analysis and identification

Abstract We describe and evaluate a model-based scheme for feature extraction and model-based signal identification which uses likelihood criteria for “edge” detection. Likelihood measures from the feature identification process are shown to provide a well behaved measure of signal interpretation confidence. We demonstrate that complex, transient signals, from one of 6 classes, can reliably be identified at signal to noise ratios of 2 and that identification does not fail until the signal to noise ratio has reached 1. Results show that the loss in identification performance resulting from the use of a heuristic, rather than an exhaustive, search strategy is minimal.

Robust statistical model-based cell image interpretation

A robust and adaptable model-based scheme for cell image interpretation is presented that can accommodate the wide natural variation in the appearance of cells. This is achieved using multiple models and an interpretation process that permits a smooth transition between the models. Boundaries are represented using trainable statistical models that are invariant to transformatio ns of scaling, shift, rotation and contrast; a Gaussian and a circular autoregressive model (CAR) are investigated. The interpretation process optimises the match between models and data using a Bayesian distance measure. We demonstrate how objects that vary in both shape and grey-level pattern can be reliably segmented. The results presented show that the overall performance is comparable with that of manual segmentation; the area within the automatically detected and the manually selected cell boundaries that is not common to both is less than 5% in 96% of the cases tested. The results also show that the computationally simpler Gaussian boundary model is at least as effective as the CAR model.

Similarity colour morphology

Mathematical morphology was developed for binary images and extended to grey-level images. To date there is no widely accepted extension of mathematical morphology to colour. We present a unifying concept for binary, grey-level and colour morphology introducing similarity measures to form classes of colour morphological operators. We define similarity criteria as the basis for mathematical morphology with flat and non-flat structuring elements. Results for dilation, erosion and hit-or-miss transforms on binary, grey-level and colour images are presented.

Multiscale medial axis through a complete set of optimal scale ridges

This paper presents an algorithm to extract a complete set of optimal scale ridges from the medialness scale space. The result is the multiscale medial axis. This algorithm differs from the previous implementations of optimal scale ridge in that it searches for the local maximum with respect to scale, rather than the global maximum. Therefore a single position is allowed to belong to separate axis segments at different scales; large scale features no longer dominate small scale features.

Managing bandwidth utilisation for image transmission

Few networks offer sufficient bandwidth for the transmission of high resolution 2-D and 3-D medical image sets without incurring significant latency. Traditional compression methods do not resolve this problem because, in most cases, the latency in decoding compressed data is similar to, or greater than, the reduction in transmission latency achieved by compression. We describe an approach to managing image transmission in which spatial regions are selected and prioritised for transmission so that data is received in a timely manner. This is a lossless form of progressive image transmission in which we first transmit an approximate, iconic form of the image. Then each part of the iconic image is progressively updated. The order in which the subparts are updated is determined using a simple set of rules. This progressive image transmission scheme takes into account the behavioural requirements of the user to make good use of the bandwidth available. This strategy addresses the need to minimise effective latency and network loading. Initial results indicate that with this scheme of bandwidth management, a valuable transmitted image is received with a latency that is about 1/10th that of other progressive image transmission schemes and that the dedicated bandwidth required to achieve this is much reduced. Frequently, the necessary regions of the image are transmitted in about half the time taken to transmit the full image.

Event detection and period extraction using multi-scale symmetry and entropy

We present a system for detecting discrete periodic events over a short interval and in the presence of interference. In the first stage symmetries are identified using scale-space representation. This process detects signal events with a low signal-to-noise ratio but has the potential to introduce a number of false responses. This process is followed by an entropy-based algorithm that can robustly extract periodicities from a set of observed discrete events in the presence of a large number of false alarms. The event detection and period extraction processes have a low computational cost and can extract signal periodicity after a short observation time. This scheme was evaluated against four previously reported methods. Results demonstrate that the period extraction algorithm presented here is more reliable than three of the previously reported algorithms. The reliability of the algorithm presented here was similar to that of the fourth method but the computational cost was much less.