In this thesis we present a method for detecting planar targets in real-time on power-constrained, or low-powered, hand-held devices such as mobile phones. We adopt the feature recognition (also referred to as feature matching) approach and employ fast-to-compute local feature descriptors to establish point correspondences. To obtain a satisfactory localization accuracy, most local feature descriptors seek a transformation of the input intensity patch that is invariant to various geometric and photometric deformations. Generally, such transformations are computationally intensive, hence are not ideal for real-time applications on limited hardware platforms. On the other hand, descriptors which are fast to compute are typically limited in their ability to provide invariance to a vast range of deformations. To address these shortcomings, we have developed a learning-based approach which can be applied to any local feature descriptor to increase the system’s robustness to both affine and perspective deformations. The motivation behind applying a learning-based approach is to transfer as much of the computational burden (as possible) onto an offline training phase, allowing a reduction in cost during online matching. The approach comprises of identifying keypoints which remain stable under artificially induced perspective transformations, extracting the corresponding feature vectors, and finally aggregating the feature vectors of coincident keypoints to obtain the final descriptors. We strictly focus on objects which are planar, thus allowing us to synthesize images of the object in order to capture the appearance of keypoint patches under several perspectives.

Recent developments in linguistic theory carried out within the Minimalist Program (Chomsky 1995; Adger 2003;) provide a functional and concrete framework for an analysis of noun phrases in the Indonesian language, a Western-Malayo Polynesian sub-branch of the Austronesian language family. An analysis of Indonesian noun phrase structure within this framework demonstrates that the head noun occurs in a base-generated position, at the bottom of a DP, while pre- and post-nominal modifiers are contained within a number of additional projections that merge above the head noun. In this thesis, the proposal is made for a relatively unrestricted adjunction analysis, whereby head adjunction via Merge allows for the direct expansion of the head N at various levels of the Indonesian DP. Evidence is presented to show that the adjoined status of attributive nouns and adjectives, a plural feature [PL], and the feature [DEF] generates a complex hierarchical structure in which there is no predefined order between a specifier or complement and the head noun. In addition, it is argued that bare nouns are neutral with respect to number and, given that number-marking, possession and (in)definiteness are optional, all projections that merge above the head N are optional and context is needed to accurately interpret an Indonesian bare noun.

Feature extraction is the most critical step in any human gait recognition system. Although gait is a dynamic process yet the static body parameters also play an important role in characterizing human gait. A few studies were performed in the past to assess the comparative relevance of static and dynamic gait features. There is, however, a lack of work in comparative performance analysis of dynamic gait features from different parts of the silhouettes in an appearance based setup. This dissertation presents a comparative study of dynamic features extracted from legs, arms and shoulders for gait recognition. Our study partially supports the general notion of leg motion being the most important determining factor in gait recognition. But it is also observed that features extracted from upper arm and shoulder area become more significant in some databases. The usefulness of the study hinges on the fact that lower parts of the leg are generally more noisy due to a variety of variations such as walking surface, occlusion and shadows. Dynamic features extracted from the upper part of the silhouettes posses significantly higher discriminatory power in such situations. In other situations these features can play a complementary role in the gait recognition process. We also propose two new feature extraction methods for gait recognition. The new methods use silhouette area signals which are easy and simple to extract. A significant performance increase is achieved by using the new features over the benchmark method and recognition results compare well to the other current techniques. The simplicity and compactness of the proposed gait features is their major advantage because it entails low computational overhead.

Feature extraction is the most critical step in any human gait recognition system. Although gait is a dynamic process yet the static body parameters also play an important role in characterizing human gait. A few studies were performed in the past to assess the comparative relevance of static and dynamic gait features. There is, however, a lack of work in comparative performance analysis of dynamic gait features from different parts of the silhouettes in an appearance based setup. This dissertation presents a comparative study of dynamic features extracted from legs, arms and shoulders for gait recognition. Our study partially supports the general notion of leg motion being the most important determining factor in gait recognition. But it is also observed that features extracted from upper arm and shoulder area become more significant in some databases. The usefulness of the study hinges on the fact that lower parts of the leg are generally more noisy due to a variety of variations such as walking surface, occlusion and shadows. Dynamic features extracted from the upper part of the silhouettes posses significantly higher discriminatory power in such situations. In other situations these features can play a complementary role in the gait recognition process. We also propose two new feature extraction methods for gait recognition. The new methods use silhouette area signals which are easy and simple to extract. A significant performance increase is achieved by using the new features over the benchmark method and recognition results compare well to the other current techniques. The simplicity and compactness of the proposed gait features is their major advantage because it entails low computational overhead.

Feature extraction is an important step in the classification of high-dimensional data such as face images. Furthermore, linear feature extractors are more prevalent due to computational efficiency and preservation of the Gaussianity. This research proposes a simple and fast linear feature extractor approximating the sufficient statistic for Gaussian distributions. This method preserves the discriminatory information in both first and second moments of the data and yields the linear discriminant analysis as a special case. Additionally, an accurate upper bound on the error probability of a plug-in classifier can be used to approximate the number of features minimizing the error probability. Therefore, tighter error bounds are derived in this work based on the Bayes error or the classification error on the trained distributions. These bounds can also be used for performance guarantee and to determine the required number of training samples to guarantee approaching the Bayes classifier performance.

Content Based Image Retrieval (CBIR) systems represent images in the database by color, texture, and shape information. In this thesis, we concentrate on tex- ture features and introduce a new generic texture descriptor, namely, Statistical Analysis of Structural Information (SASI). Moreover, in order to increase the re- trieval rates of a CBIR system, we propose a new method that can also adapt an image retrieval system into a con&macr / gurable one without changing the underlying feature extraction mechanism and the similarity function. SASI is based on statistics of clique autocorrelation coe&plusmn / cients, calculated over structuring windows. SASI de&macr / nes a set of clique windows to extract and measure various structural properties of texture by using a spatial multi- resolution method. Experimental results, performed on various image databases, indicate that SASI is more successful then the Gabor Filter descriptors in cap- turing small granularities and discontinuities such as sharp corners and abrupt changes. Due to the &deg / exibility in designing the clique windows, SASI reaches higher average retrieval rates compared to Gabor Filter descriptors. However, the price of this performance is increased computational complexity. Since, retrieving of similar images of a given query image is a subjective task, it is desirable that retrieval mechanism should be con&macr / gurable by the user. In the proposed method, basically, original feature space of a content-based retrieval system is nonlinearly transformed into a new space, where the distance between the feature vectors is adjusted by learning. The transformation is realized by Arti&macr / cial Neural Network architecture. A cost function is de&macr / ned for learning and optimized by simulated annealing method. Experiments are done on the texture image retrieval system, which use SASI and Gabor Filter features. The results indicate that con&macr / gured image retrieval system is signi&macr / cantly better than the original system.

Some insects have the capability to detect and track small moving objects, often against cluttered moving backgrounds. Determining how this task is performed is an intriguing challenge, both from a physiological and computational perspective. Previous research has characterized higher-order neurons within the fly brain known as 'small target motion detectors‘ (STMD) that respond selectively to targets, even within complex moving surrounds. Interestingly, these cells still respond robustly when the velocity of the target is matched to the velocity of the background (i.e. with no relative motion cues). We performed intracellular recordings from intermediate-order neurons in the fly visual system (the medulla). These full-wave rectifying, transient cells (RTC) reveal independent adaptation to luminance changes of opposite signs (suggesting separate 'on‘ and 'off‘ channels) and fast adaptive temporal mechanisms (as seen in some previously described cell types). We show, via electrophysiological experiments, that the RTC is temporally responsive to rapidly changing stimuli and is well suited to serving an important function in a proposed target-detecting pathway. To model this target discrimination, we use high dynamic range (HDR) natural images to represent 'real-world‘ luminance values that serve as inputs to a biomimetic representation of photoreceptor processing. Adaptive spatiotemporal high-pass filtering (1st-order interneurons) shapes the transient 'edge-like‘ responses, useful for feature discrimination. Following this, a model for the RTC implements a nonlinear facilitation between the rapidly adapting, and independent polarity contrast channels, each with centre-surround antagonism. The recombination of the channels results in increased discrimination of small targets, of approximately the size of a single pixel, without the need for relative motion cues. This method of feature discrimination contrasts with traditional target and background motion-field computations. We show that our RTC-based target detection model is well matched to properties described for the higher-order STMD neurons, such as contrast sensitivity, height tuning and velocity tuning. The model output shows that the spatiotemporal profile of small targets is sufficiently rare within natural scene imagery to allow our highly nonlinear 'matched filter‘ to successfully detect many targets from the background. The model produces robust target discrimination across a biologically plausible range of target sizes and a range of velocities. We show that the model for small target motion detection is highly correlated to the velocity of the stimulus but not other background statistics, such as local brightness or local contrast, which normally influence target detection tasks. From an engineering perspective, we examine model elaborations for improved target discrimination via inhibitory interactions from correlation-type motion detectors, using a form of antagonism between our feature correlator and the more typical motion correlator. We also observe that a changing optimal threshold is highly correlated to the value of observer ego-motion. We present an elaborated target detection model that allows for implementation of a static optimal threshold, by scaling the target discrimination mechanism with a model-derived velocity estimation of ego-motion. Finally, we investigate the physiological relevance of this target discrimination model. We show that via very subtle image manipulation of the visual stimulus, our model accurately predicts dramatic changes in observed electrophysiological responses from STMD neurons. / http://proxy.library.adelaide.edu.au/login?url= http://library.adelaide.edu.au/cgi-bin/Pwebrecon.cgi?BBID=1368818 / Thesis (Ph.D.) - University of Adelaide, School of Molecular and Biomedical Science, 2009

“Machine learning is the process of discovering and interpreting meaningful information, such as new correlations, patterns and trends by sifting through large amounts of data stored in repositories, using pattern recognition technologies as well as statistical and mathematical techniques” (Larose, 2005). From my understanding, machine learning is a process of using different analysis techniques to observe previously unknown, potentially meaningful information, and discover strong patterns and relationships from a large dataset. Professor Kasabov (2007b) classified computational models into three categories (e.g. global, local, and personalised) which have been widespread and used in the areas of data analysis and decision support in general, and in the areas of medicine and bioinformatics in particular. Most recently, the concept of personalised modelling has been widely applied to various disciplines such as personalised medicine, personalised drug design for known diseases (e.g. cancer, diabetes, brain disease, etc.) as well as for other modelling problems in ecology, business, finance, crime prevention, and so on. The philosophy behind the personalised modelling approach is that every person is different from others, thus he/she will benefit from having a personalised model and treatment. However, personalised modelling is not without issues, such as defining the correct number of neighbours or defining an appropriate number of features. As a result, the principal goal of this research is to study and address these issues and to create a novel framework and system for personalised modelling. The framework would allow users to select and optimise the most important features and nearest neighbours for a new input sample in relation to a certain problem based on a weighted variable distance measure in order to obtain more precise prognostic accuracy and personalised knowledge, when compared with global modelling and local modelling approaches.

“Machine learning is the process of discovering and interpreting meaningful information, such as new correlations, patterns and trends by sifting through large amounts of data stored in repositories, using pattern recognition technologies as well as statistical and mathematical techniques” (Larose, 2005). From my understanding, machine learning is a process of using different analysis techniques to observe previously unknown, potentially meaningful information, and discover strong patterns and relationships from a large dataset. Professor Kasabov (2007b) classified computational models into three categories (e.g. global, local, and personalised) which have been widespread and used in the areas of data analysis and decision support in general, and in the areas of medicine and bioinformatics in particular. Most recently, the concept of personalised modelling has been widely applied to various disciplines such as personalised medicine, personalised drug design for known diseases (e.g. cancer, diabetes, brain disease, etc.) as well as for other modelling problems in ecology, business, finance, crime prevention, and so on. The philosophy behind the personalised modelling approach is that every person is different from others, thus he/she will benefit from having a personalised model and treatment. However, personalised modelling is not without issues, such as defining the correct number of neighbours or defining an appropriate number of features. As a result, the principal goal of this research is to study and address these issues and to create a novel framework and system for personalised modelling. The framework would allow users to select and optimise the most important features and nearest neighbours for a new input sample in relation to a certain problem based on a weighted variable distance measure in order to obtain more precise prognostic accuracy and personalised knowledge, when compared with global modelling and local modelling approaches.

Zugl.: Aachen, Techn. Hochsch., Diss., 2007.