No description available.

No description available.

This thesis presents a new algorithm which can be used to register images of the same or different modalities e.g images with multiple channels such as X-rays, temperature, or elevation or simply images of different spectral bands. In particular, a correlation-based scheme is used, but instead of grey values, it correlates numbers formulated by different combinations of the extracted local Walsh coefficients of the images. Each image patch is expanded in terms of Walsh basis functions. Each Walsh basis function can be thought of as measuring a different aspect of local structure, eg horizontal edge, corner, etc. The coefficients of the expansion, therefore, can be thought of as dense local features, estimating at each point the degree of presence of, for example, a horizontal edge, a corner with contrast of a certain type, etc. These coefficients are normalised and used as digits in a chosen number system which allows one to create a unique number for each type of local structure. The choice of the basis of the number system allows one to give different emphasis to different types of local feature (e.g. corners versus edges) and thus the method we present forms a unified framework in terms of which several feature matching methods may be interpreted. The algorithm is compared with wavelet based approaches, using simulated and real images. The images used for the registration experiments are assumed to differ from each other by a rotation and a translation only. Additionally, the method was extended to cope with 3D image sets, while as an add-on, it was also tried in performing image segmentation.

Changes in the illumination condition cause dramatic variation in face appearance and seriously affect the performance of face recognition systems. This problem is addressed in the thesis by introducing an approach based on active Near-Infrared differential imaging. Assuming a static scene, a linear response of the sensor to the scene radiation and no saturation, it is shown theoretically and empirically in this thesis that the active differential imaging technique yields a face image independent of the change in ambient illumination. By taking the difference of two face images, one captured with the active illumination on and one with it off, the resulting image contains the face illuminated only by the active illumination source. This technique is compared with several representative illumination invariant face recognition techniques on a database containing faces captured under different illuminations and at different time. The results in face identification and verification experiments demonstrate the significant advantage of this Near-Infrared differential imaging technique over the other techniques. This thesis also presents a multistage approach to automatic face localisation for the Near-Infrared face images. This multistage approach is a combination of a novel pupil detection approach based on edge following and chaincode representation, and an approach based on FloatBoost learning. Accurate face localisation results are achieved by the proposed multistage approach, and this leads to the excellent face recognition performance in fully automatic scenario. A subject appears at two different locations before and after the active illumination is turned on if he/she is moving. This causes motion artifact in the difference image from the active differential imaging system and degrades the performance of the face recognition system. The thesis presents an approach based on motion compensation to deal with this problem. It is shown from the experimental results that the proposed approach successfully removes the motion artifacts and improves the face recognition performance significantly.

Structured Light Scanners measure the surface of a target object, producing a set of vertices which can be used to construct a three-dimensional model of the surface. The techniques are particularly appropriate for measuring the smoothly undulating, featureless forms which Stereo Vision methods find difficult, and the structured light pattern explicitly gives a dense graph of connected vertices, thus obviating the need for vertex-triangulation prior to surface reconstruction. In addition, the technique provides the measurements almost instantaneously, and so is suitable for scanning moving and non-rigid objects. Because of these advantages there is an imperative to extend the range of scannable surfaces to those including occlusions, which often reduce or prevent successful measurement. This thesis investigates ways of improving both the accuracy and the range of surface types which can be scanned using structured light techniques, extending current research by examining the role of occlusions and geometric constraints, and introducing novel algorithms to solve the Indexing Problem. The Indexing Problem demands that for every pattern element in the projected image, its counterpart, reflected from the surface of the target object, must be found in the recorded image, and most researchers have declared this problem to be intractable without resorting to coding schemes which uniquely identify each pattern element. The use of uncoded projection patterns, where the pattern elements are projected without any unique identification, has two advantages: firstly it provides the densest possible set of measured vertices within a single video timeframe, and secondly it allows the investigation of the fundamental problems without the distraction of dealing with coding schemes. These advantages educe the general strategy adopted in this thesis, of attempting to solve the Indexing Problem using uncoded patterns, and then adding some coding where difficulties still remain. In order to carry out these investigations it is necessary to precisely measure the system and its outputs, and to achieve this requirement two scanners have been built, a Single Stripe Scanner and a Multiple Stripe Scanner. The Single Stripe Scanner introduces the geometric measurement methods and provides a reference output which matches the industry standard; the Multiple Stripe Scanner then tests the results of the investigations and evaluates the success of the new algorithms and constraints. In addition, some of the investigations are tested theoretically, by using synthetic data and by the solution of geometric diagrams. These evaluations of success show that, if occlusions are not present in the recorded data, the Indexing Problem can often be completely solved if the new indexing algorithms and geometric constraints are included. Furthermore, while there are some cases where the Indexing Problem cannot be solved without recourse to a coding scheme, the addition of occlusion detection in the algorithms greatly improves the indexing accuracy and therefore the successful measurement of the target surface.

Live iris image analysis based personal identification has received more and more attention in today's highly mobile and inter-connected society, with deteriorating situation in public security. Huge amount of work has been done and great progress achieved in this area. However, some critical problems still persist and significant work needs to be done before mass-scale deployment on national and international levels can be achieved. Recognition performance and the system speed are the two hot topics in iris recognition research. The research aim in this thesis is trying to answer the following two questions: (1) what kind of information within the iris textures could be utilized for authentication? (2) how the algorithms could be speeded up for real time system requirement? The state of the art in iris recognition technologies is reviewed at first and the research difficulties are pointed out. In order to carry out the research in Bath, the Bath Iris Recognition Research Environment is built based on extensive literature review and comparison. Then four algorithms are proposed with the aim of answering the two questions above. A fast and robust iris localization algorithm based on Random Sample Consensus (RANSAC) is proposed in Chapter four, which could not only maintain a high localization rate, but also perform the procedure much faster than the leading algorithms proposed by Daugman or Wildes. This work contributes to question 2. Chapter four also reported a fast and robust eyelid removal algorithm, which could be carried out at a very high speed while still keeping a satisfying correct removal rate. This work also contributes to question 2. Chapter five proposed a local frequency amplitude variation based iris coding algorithm, which could greatly reduce the processing time while still maintain a very high distinguishing capability. This work contributes to both question 1 and 2. An effective eyelash removal algorithm based on local area analysis has been proposed in chapter six. Unlike other previous eyelash removal methods, which generally tried to detect and mask the eyelashes or the eyelash areas, the proposed method recreates iris pixels occluded by eyelashes using information from their non-occluded neighbors. Extensive experiments and comparisons have been done to prove the effectiveness and low computation complexity of these four proposed algorithms. Through the research in this thesis, two possible answers to the questions proposed at the beginning of this thesis could be given: (1) Local image variation is the essential unique information for one iris class to be differentiated from another. Apart from utilizing the local phase or intensity variations, this thesis used the local frequency variation, which is also proved to be very effective. Statistically, the local iris image pixels are related to each other, which could be used to reconstruct occluded iris pixels. (2) The adoption of Fast Fourier Transform (FFT) makes the feature extraction procedure very time efficient. Thus the fast representation of local feature information is a very effective way to speed up the coding process. Also the adoption of fast iris image preprocessing techniques would contribute to the system speed.

No description available.

The satellite imaging service used to be very expensive and mainly explored by government for civil and military applications. With the development and commercialisation of space industry, there is increasing demand on the timely satellite imagery service. Therefore the performance analysis of the service delivery delay of the satellite imaging system becomes important. Currently the performance of satellite imaging service is mostly analysed based on system simulation results or real sample data from service providers, which are usually accurate but not systematic. Also the effects of system parameters are not direct viewing. Theoretical modelling approach, however, captures the feature of the system and provides direct systematic performance evaluation, which is a useful guide in designing satellite system according to the user's requirements in timely service delivery. Therefore this work is motivated to model the satellite imaging system analytically. In this work we consider a satellite imaging system with single satellite, single ground station and limited onboard storage. Queueing theory is applied for the performance modelling. Since there are two tandem service stages: image capture and image download, the system is modelled using a two-node tandem queue with blocking. At first we propose a continuous-time two-dimensional Markov chain for describing the satellite imaging system. The model works well but has its limitations and disadvantages. In order to have a better model for the satellite imaging system, we conduct a further research on the service mechanism of the image capture and image download. Using satellite geometric models, we derive the stochastic features of image capture and download service, in terms of the probability of having an operation opportunity each orbital period, and the probability distribution of the inter-time between operation opportunities. We propose discrete-time clocking models for the image capture and download service respectively, with queue status defined on a discrete-time orbit- by-orbit basis, and with transition probabilities between system states derived from the stochastic analysis. We then concatenate these two queues and propose a tandem clocking model for the complete satellite imaging service system. The discrete-time clocking models have been evaluated by comparison with simulation results using different target location distributions. The models are proven to be very successful for general distributed target locations, with above 95% high accuracy for the mean system queue length. Using the tandem clocking model, we conduct performance analysis for the key service performance measure: average waiting time, and also investigate system parameters' effects on the system performance. With this work, we are enabled to provide information about the general performance of a satellite imaging system, which caters to the customer's concern and also is very helpful for system optimisation in satellite design and control.

This thesis is on the use of a surface reconstruction method called Photometric Stereo in Machine Vision. In the first part we propose a modification of the method which allows one to reconstruct the shape and colour of the surface in the presence of highlights and shadows. We also provide an extensive error analysis of the algorithm, concentrating on design issues such as choice of thresholds. The second part of the thesis is on the subject of illumination direction invariant classification of 3-dimensional textures. Using Photometric Stereo, one can reconstruct the surface shape and infer its statistical properties. The learned statistics can be used to predict the behaviour of a class of texture features, which in its turn allows one to recognise a 3-dimensional texture under unknown illumination.

In recent years, the performance of scientific CMOS active pixel sensors has been improved to the point that it is now approaching that of the current silicon sensor of choice, CCDs. For some applications, CMOS APSs is believed to present significant advantages over CCDs, such as improved radiation hardness. In this work, the effect of radiation damage on a ‘baseline’ commercial APS, e2v technologies’ Jade APS, is characterised in response to gamma, proton and heavy ion irradiation. Specific performance problems encountered during this radiation characterisation, such as dark current non-uniformity under gamma irradiation, random telegraph signals under proton irradiation, and single event effects under heavy ion irradiation are described and analyzed. The X-ray spectroscopic imaging performance of the device is measured and compared to the Ocean Colour Imager APS test array showing progress towards a high frame rate spectroscopic X-ray imager for space science. The implications of these results for using similar devices in space applications are considered. Furthermore, possible novel techniques for measuring inter-pixel responsivity non-uniformity, heavy ion detection and spectroscopy, and measuring the dynamics of radiation-induced trap formation are discussed.