The outcome of this study clearly shows the applicability of this hybrid method in the analysis of skin features and is therefore expected to lead development of non-invasive and low-cost instrument for early detection of skin changes.

No abstract

Objective: The textural structure of "skin age" related sub-skin components enables us to identify and analyse their unique characteristics, thus making substantial progress towards establishing an accurate skin age model. Methods: This is achieved by a two stage process. First by the application of textural analysis using laser speckle imaging, which is sensitive to textural effects within the λ=650 nm spectral band region. In the second stage a Bayesian inference method is used to select attributes from which a predictive model is built. Results: This technique enables us to contrast different skin age models, such as the laser-speckle effect against the more widely used normal light (LED) imaging method, whereby it is shown that our laser speckle based technique yields better results. Conclusion: The method introduced here is non-invasive, low-cost and capable of operating in real-time; having the potential to compete against high-cost instrumentation such as confocal microscopy or similar imaging devices used for skin age identification purposes.

The recent increase in ageing population in countries around the world has brought a lot of attention toward research and development of ambient assisted living (AAL) systems. These systems should be inexpensive to be installed in elderly homes, protecting their privacy and more importantly being non-invasive and smart. In this paper, we introduce an inexpensive system that utilises off-the-shelf sensor to grab RGB-D data. This data is then fed into different learning algorithms for classification different activity types. We achieve a very good success rate (99.9%) for human activity recognition (HAR) with the help of light-weighted and fast random forests (RF).

A classification technique which distinguishes between manmade and natural textural features visible on high resolution satellite images is introduced. The proposed work aims to evaluate non-linear classification techniques by the unification of appropriate texture analysis methods and a learningBayesian classifier which is more robust against data uncertainty than the other types of linear classifiers. The classification technique introduced within this work will also provide an opportunity for fully automated thematic and land-use map generation.

Purpose : As a consequence of the latest developments in laser technologies it is now possible to develop a low-cost and accurate tablet inspection system by the unification of optical and artificial intelligence methods. Method: The functionality of the proposed system is based on a sequence of texture analysis of laser speckle images (using laser sources of 650 nm and 808 nm : VIS/IR) followed by the optimization of texture parameters using Bayesian Networks (BN). Results: In the first part of this work, a Bayesian inference method was used to detect micro-scale tablet defects that are generated "progressively" during production whereas in the second part a Bayesian classifier method was used to discriminate between tablets made from different granule sizes. In part two, it was shown that (i) the comparatively higher energy (5mW) IR laser light generates different speckle effects than the lower energy visible (Red 3mW) by interacting with deeper sub-surface of the tablets and (ii) by using multiclassifier systems (MCS) to fuse the Bayesian classifiers from both types of speckle images it was possible to achieve a higher discrimination power (88% classification accuracy) for distinguishing between tablets made from different granule sizes than one can achieve from a single image type. Conclusion: It is suggested that this unified method has the potential to provide for a comprehensive analysis of both tablet quality attributes, on the one hand, and failure modes, on the other, that might be used in the development of a low cost tablet inspection system.

No abstract

The method proposed here uses Bayesian non-linear classifier to select optimal subset of attributes to avoid redundant variables and reduce data uncertainty in the classification process often used in medical diagnosis. The method also exploits the structural reasoning ability of Bayesian Networks (BN) to optimize large number of attributes to prevent overfitting, meanwhile it maintains the high classification accuracy. This process simplifies the complex data analyses and may lead to a cost reduction in clinical data acquisition process.