K. C. P. Wong

DARBS: a distributed blackboard system

Prior to this work, an algorithmic and rule-based blackboard system (ARBS) had been developed over a ten-year period. ARBS benefited from a versatile rule structure and the ability to mix computational styles either as separate knowledgesources or by embedding algorithms within rules. It was a serial system – any knowledge source that was able to contribute had to wait its turn. We report here on a new distributed system, DARBS, in which the knowledge sources are parallel processes. Based around the client/server model, DARBS comprises a centralised database server, i.e. the blackboard, and a number of knowledge source clients. As the clients are separate processes, possibly on separate networked computers, they can contribute to the solution of a problem whenever they have a contribution to make. DARBS therefore achieves the well-established but elusive ideal of opportunism. It behaves as a distributed agent-based system, with the proviso that all communication is via the blackboard. DARBS is currently being applied to automatic interpretation of nondestructive evaluation (NDE) data and control of plasma deposition processes.

Multimodal retinal image registration using a fast principal component analysis hybrid-based similarity measure

Multimodal retinal images (RI) are extensively used for analysing various eye diseases and conditions such as myopia and diabetic retinopathy. The incorporation of either two or more RI modalities provides complementary structure information in the presence of non-uniform illumination and low-contrast homogeneous regions. It also presents significant challenges for retinal image registration (RIR). This paper investigates how the Expectation Maximization for Principal Component Analysis with Mutual Information (EMPCA-MI) algorithm can effectively achieve multimodal RIR. This iterative hybrid-based similarity measure combines spatial features with mutual information to provide enhanced registration without recourse to either segmentation or feature extraction. Experimental results for clinical multimodal RI datasets comprising colour fundus and scanning laser ophthalmoscope images confirm EMPCA-MI is able to consistently afford superior numerical and qualitative registration performance compared with existing RIR techniques, such as the bifurcation structures method.

Robust retinal image registration using expectation maximisation with mutual information

Retinal images (RI) are widely used to diagnose a variety of eye conditions and diseases such as myopia and diabetic retinopathy. They are inherently characterised by having nonuniform illumination and low-contrast homogeneous regions which represent a unique set of challenges for retinal image registration (RIR). This paper investigates using the expectation maximization for principal component analysis based mutual information (EMPCA-MI) algorithm in RIR. It combines spatial features with mutual information to efficiently achieve improved registration performance. Experimental results for mono-modal RI datasets verify that EMPCA-MI together with Powell-Brent optimization affords superior robustness in comparison with existing RIR methods, including the geometrical features method.

New dynamic spectrum access algorithm for TV white space cognitive radio networks

As more user applications emerge for wireless devices, the corresponding amount of traffic is rapidly expanding, with the corollary that ever-greater spectrum capacity is required. Service providers are experiencing deployment blockages due to insufficient bandwidth being available to accommodate such devices. TV White Space (TVWS) represents an opportunity to supplement existing licensed spectrum by exploiting unlicensed resources. TVWS spectrum has materialised from the unused TV channels in the switchover from analogue to digital platforms. The main obstacles to TVWS adoption are reliable detection of primary users (PU) i.e., TV operators and consumers, allied with specifically, the hidden node problem. This paper presents a new Generalised Enhanced Detection Algorithm (GEDA) that exploits the unique way Digital Terrestrial TV (DTT) channels are deployed in different geographical areas. GEDA effectively transforms an energy detector into a feature sensor to achieve significant improvements in detection probability of a DTT PU. Furthermore, by framing a novel margin strategy utilising a keep out contour, the hidden node issue is resolved and a viable secondary user sensing solution formulated. Experimental results for a cognitive radio TVWS model have formalised both the bandwidth and throughput gains secured by TVWS users with this new paradigm.

High-motion table tennis ball tracking for umpiring applications

Table-tennis umpiring presents many challenges where technology can be judiciously applied to enhance decision-making, especially in the service facet of the game. This paper presents a system to automatically detect and track the ball during table-tennis services to enable precise judgment over their legitimacy. The system comprises a suite of algorithms that adaptively exploit spatial and temporal information from real match videos, which are generally characterized by high object motion, allied with object blurring and occlusion. Experimental results on various table-tennis test videos corroborate the system performance in facilitating accurate and efficient decision-making over the validity of a service.

A new cross-layer design strategy for TV White Space Cognitive Radio applications

Conventional single layer processing in Cognitive Radio Networks (CRN) can incur significant time costs in transferring information between the various layers of the Open System Interconnection (OSI) model, due its innate sequential structure. This is especially a problem for CRN which usually only has a narrow time window to access spectral gaps of either licensed or other secondary users (SU). To exploit this opportunity, cross layer processing (CLP) paradigms that share information between OSI layers and the radio system have been proposed to maximise throughput for SU, while maintaining Quality of Service (QoS) provision to the licensed primary user. With the global transference of TV systems to digital platforms, regulatory bodies have identified an opportunity to allocate additional digital TV (DTV) channels to CRNs on a localised basis, in what is called TV White Space (TVWS). This paper investigates how CLP of information can be effectively exploited to enhance CRN system performance by making key channel allocations to minimise disruption to the spectrum environment, while maximising available resources to fulfil application and network requirements within TVWS.

Enhanced retinal image registration accuracy using expectation maximisation and variable bin-sized mutual information

While retinal images (RI) assist in the diagnosis of various eye conditions and diseases such as glaucoma and diabetic retinopathy, their innate features including low contrast homogeneous and non-uniformly illuminated regions, present a particular challenge for retinal image registration (RIR). Recently, the hybrid similarity measure, Expectation Maximization for Principal Component Analysis with Mutual Information (EMPCA-MI) has been proposed for RIR. This paper investigates incorporating various fixed and adaptive bin size selection strategies to estimate the probability distribution in the mutual information (MI) stage of EMPCA-MI, and analyses their corresponding effect upon RIR performance. Experimental results using a clinical mono-modal RI dataset confirms that adaptive bin size selection consistently provides both lower RIR errors and superior robustness compared to the empirically determined fixed bin sizes.

Fast EM Principal Component Analysis Image Registration Using Neighbourhood Pixel Connectivity

Image registration IR is the systematic process of aligning two images of the same or different modalities. The registration of mono and multimodal images i.e., magnetic resonance images, pose a particular challenge due to intensity non-uniformities INU and noise artefacts. Recent similarity measures including regional mutual information RMI and expectation maximisation for principal component analysis with MI EMPCA-MI have sought to address this problem. EMPCA-MI incorporates neighbourhood region information to iteratively compute principal components giving superior IR performance compared with RMI, though it is not always effective in the presence of high INU. This paper presents a modified EMPCA-MI mEMPCA-MI similarity measure which introduces a novel pre-processing step to exploit local spatial information using 4-and 8-pixel neighbourhood connectivity. Experimental results using diverse image datasets, conclusively demonstrate the improved IR robustness of mEMPCA-MI when adopting second-order neighbourhood representations. Furthermore, mEMPCA-MI with 4-pixel connectivity is notably more computationally efficient than EMPCA-MI.

Disocclusion hole-filling in DIBR-synthesized images using multi-scale template matching

Transmitting texture and depth images of captured camera view(s) of a 3D scene enables a receiver to synthesize novel virtual viewpoint images via Depth-Image-Based Rendering (DIBR). However, a DIBR-synthesized image often contains disocclusion holes, which are spatial regions in the virtual view image that were occluded by foreground objects in the captured camera view(s). In this paper, we propose to complete these disocclusion holes by exploiting the self-similarity characteristic of natural images via nonlocal template-matching (TM). Specifically, we first define self-similarity as nonlocal recurrences of pixel patches within the same image across different scales-one characterization of self-similarity in a given image is the scale range in which these patch recurrences take place. Then, at encoder we segment an image into multiple depth layers using available per-pixel depth values, and characterize self-similarity in each layer with a scale range; scale ranges for all layers are transmitted as side information to the decoder. At decoder, disocclusion holes are completed via TM on a per-layer basis by searching for similar patches within the designated scale range. Experimental results show that our method improves the quality of rendered images over previous disocclusion hole-filling algorithms by up to 3.9dB in PSNR.

Robust image registration using adaptive expectation maximisation based PCA

Images having either the same or different modalities can be aligned using the systematic process of image registration. Inherent image characteristics including intensity non-uniformities in magnetic resonance images and large homogeneous non-vascular regions in retinal and other generic image types however, pose a significant challenge to their registration. This paper presents an adaptive expectation maximisation for principal component analysis with mutual information (aEMPCA-MI) similarity measure for image registration. It introduces a novel iterative process to adaptively select the most significant principal components using Kaiser rule and applies 4-pixel connectivity for feature extraction together with Wichards bin size selection in calculating the MI. Both quantitative and qualitative results on a diverse range of image datasets, conclusively demonstrate the superior image registration performance of aEMPCA-MI compared with existing Mi-based similarity measures.