Pattern recognition using invariants defined from higher order spectra : 2-D image inputs

A new algorithm for extracting features from images for object recognition is described. The algorithm uses higher order spectra to provide desirable invariance properties, to provide noise immunity, and to incorporate nonlinearity into the feature extraction procedure thereby allowing the use of simple classifiers. An image can be reduced to a set of 1D functions via the Radon transform, or alternatively, the Fourier transform of each 1D projection can be obtained from a radial slice of the 2D Fourier transform of the image according to the Fourier slice theorem. A triple product of Fourier coefficients, referred to as the deterministic bispectrum, is computed for each 1D function and is integrated along radial lines in bifrequency space. Phases of the integrated bispectra are shown to be translation- and scale-invariant. Rotation invariance is achieved by a regrouping of these invariants at a constant radius followed by a second stage of invariant extraction. Rotation invariance is thus converted to translation invariance in the second step. Results using synthetic and actual images show that isolated, compact clusters are formed in feature space. These clusters are linearly separable, indicating that the nonlinearity required in the mapping from the input space to the classification space is incorporated well into the feature extraction stage. The use of higher order spectra results in good noise immunity, as verified with synthetic and real images. Classification of images using the higher order spectra-based algorithm compares favorably to classification using the method of moment invariants

Digit recognition using trispectral features

Features derived from the trispectra of DFT magnitude slices are used for multi-font digit recognition. These features are insensitive to translation, rotation, or scaling of the input. They are also robust to noise. Classification accuracy tests were conducted on a common data base of 256× 256 pixel bilevel images of digits in 9 fonts. Randomly rotated and translated noisy versions were used for training and testing. The results indicate that the trispectral features are better than moment invariants and affine moment invariants. They achieve a classification accuracy of 95% compared to about 81% for Hu's (1962) moment invariants and 39% for the Flusser and Suk (1994) affine moment invariants on the same data in the presence of 1% impulse noise using a 1-NN classifier. For comparison, a multilayer perceptron with no normalization for rotations and translations yields 34% accuracy on 16× 16 pixel low-pass filtered and decimated versions of the same data.

Characterizing the signal pattern of a four-cylinder diesel engine using acoustic emission and vibration analysis

Condition monitoring of diesel engines can prevent unpredicted engine failures and the associated consequence. This paper presents an experimental study of the signal characteristics of a 4-cylinder diesel engine under various loading conditions. Acoustic emission, vibration and in-cylinder pressure signals were employed to study the effectiveness of these techniques for condition monitoring and identifying symptoms of incipient failures. An event driven synchronous averaging technique was employed to average the quasi-periodic diesel engine signal in the time domain to eliminate or minimize the effect of engine speed and amplitude variations on the analysis of condition monitoring signal. It was shown that acoustic emission (AE) is a better technique than vibration method for condition monitor of diesel engines due to its ability to produce high quality signals (i.e., excellent signal to noise ratio) in a noisy diesel engine environment. It was found that the peak amplitude of AE RMS signals correlating to the impact-like combustion related events decreases in general due to a more stable mechanical process of the engine as the loading increases. A small shift in the exhaust valve closing time was observed as the engine load increases which indicates a prolong combustion process in the cylinder (to produce more power). On the contrary, peak amplitudes of the AE RMS attributing to fuel injection increase as the loading increases. This can be explained by the increase fuel friction caused by the increase volume flow rate during the injection. Multiple AE pulses during the combustion process were identified in the study, which were generated by the piston rocking motion and the interaction between the piston and the cylinder wall. The piston rocking motion is caused by the non-uniform pressure distribution acting on the piston head as a result of the non-linear combustion process of the engine. The rocking motion ceased when the pressure in the cylinder chamber stabilized.

A multi-resolution image alignment technique based on direct methods for pose estimation of aerial vehicles

In this paper, we seek to expand the use of direct methods in real-time applications by proposing a vision-based strategy for pose estimation of aerial vehicles. The vast majority of approaches make use of features to estimate motion. Conversely, the strategy we propose is based on a MR (Multi- Resolution) implementation of an image registration technique (Inverse Compositional Image Alignment ICIA) using direct methods. An on-board camera in a downwards-looking configuration, and the assumption of planar scenes, are the bases of the algorithm. The motion between frames (rotation and translation) is recovered by decomposing the frame-to-frame homography obtained by the ICIA algorithm applied to a patch that covers around the 80% of the image. When the visual estimation is required (e.g. GPS drop-out), this motion is integrated with the previous known estimation of the vehicles’ state, obtained from the on-board sensors (GPS/IMU), and the subsequent estimations are based only on the vision-based motion estimations. The proposed strategy is tested with real flight data in representative stages of a flight: cruise, landing, and take-off, being two of those stages considered critical: take-off and landing. The performance of the pose estimation strategy is analyzed by comparing it with the GPS/IMU estimations. Results show correlation between the visual estimation obtained with the MR-ICIA and the GPS/IMU data, that demonstrate that the visual estimation can be used to provide a good approximation of the vehicle’s state when it is required (e.g. GPS drop-outs). In terms of performance, the proposed strategy is able to maintain an estimation of the vehicle’s state for more than one minute, at real-time frame rates based, only on visual information.

Scene invariant crowd counting and crowd occupancy analysis

In public places, crowd size may be an indicator of congestion, delay, instability, or of abnormal events, such as a fight, riot or emergency. Crowd related information can also provide important business intelligence such as the distribution of people throughout spaces, throughput rates, and local densities. A major drawback of many crowd counting approaches is their reliance on large numbers of holistic features, training data requirements of hundreds or thousands of frames per camera, and that each camera must be trained separately. This makes deployment in large multi-camera environments such as shopping centres very costly and difficult. In this chapter, we present a novel scene-invariant crowd counting algorithm that uses local features to monitor crowd size. The use of local features allows the proposed algorithm to calculate local occupancy statistics, scale to conditions which are unseen in the training data, and be trained on significantly less data. Scene invariance is achieved through the use of camera calibration, allowing the system to be trained on one or more viewpoints and then deployed on any number of new cameras for testing without further training. A pre-trained system could then be used as a ‘turn-key’ solution for crowd counting across a wide range of environments, eliminating many of the costly barriers to deployment which currently exist.

Applications of medical image processing in the diagnosis and treatment of spinal deformity

In Chapter 10, Adam and Dougherty describe the application of medical image processing to the assessment and treatment of spinal deformity, with a focus on the surgical treatment of idiopathic scoliosis. The natural history of spinal deformity and current approaches to surgical and non-surgical treatment are briefly described, followed by an overview of current clinically used imaging modalities. The key metrics currently used to assess the severity and progression of spinal deformities from medical images are presented, followed by a discussion of the errors and uncertainties involved in manual measurements. This provides the context for an analysis of automated and semi-automated image processing approaches to measure spinal curve shape and severity in two and three dimensions.

Graph rigidity for near-coplanar structure from motion

Recent algorithms for monocular motion capture (MoCap) estimate weak-perspective camera matrices between images using a small subset of approximately-rigid points on the human body (i.e. the torso and hip). A problem with this approach, however, is that these points are often close to coplanar, causing canonical linear factorisation algorithms for rigid structure from motion (SFM) to become extremely sensitive to noise. In this paper, we propose an alternative solution to weak-perspective SFM based on a convex relaxation of graph rigidity. We demonstrate the success of our algorithm on both synthetic and real world data, allowing for much improved solutions to marker less MoCap problems on human bodies. Finally, we propose an approach to solve the two-fold ambiguity over bone direction using a k-nearest neighbour kernel density estimator.

Multifractal characterisation and analysis of complex networks

Complex networks have been studied extensively due to their relevance to many real-world systems such as the world-wide web, the internet, biological and social systems. During the past two decades, studies of such networks in different fields have produced many significant results concerning their structures, topological properties, and dynamics. Three well-known properties of complex networks are scale-free degree distribution, small-world effect and self-similarity. The search for additional meaningful properties and the relationships among these properties is an active area of current research. This thesis investigates a newer aspect of complex networks, namely their multifractality, which is an extension of the concept of selfsimilarity. The first part of the thesis aims to confirm that the study of properties of complex networks can be expanded to a wider field including more complex weighted networks. Those real networks that have been shown to possess the self-similarity property in the existing literature are all unweighted networks. We use the proteinprotein interaction (PPI) networks as a key example to show that their weighted networks inherit the self-similarity from the original unweighted networks. Firstly, we confirm that the random sequential box-covering algorithm is an effective tool to compute the fractal dimension of complex networks. This is demonstrated on the Homo sapiens and E. coli PPI networks as well as their skeletons. Our results verify that the fractal dimension of the skeleton is smaller than that of the original network due to the shortest distance between nodes is larger in the skeleton, hence for a fixed box-size more boxes will be needed to cover the skeleton. Then we adopt the iterative scoring method to generate weighted PPI networks of five species, namely Homo sapiens, E. coli, yeast, C. elegans and Arabidopsis Thaliana. By using the random sequential box-covering algorithm, we calculate the fractal dimensions for both the original unweighted PPI networks and the generated weighted networks. The results show that self-similarity is still present in generated weighted PPI networks. This implication will be useful for our treatment of the networks in the third part of the thesis. The second part of the thesis aims to explore the multifractal behavior of different complex networks. Fractals such as the Cantor set, the Koch curve and the Sierspinski gasket are homogeneous since these fractals consist of a geometrical figure which repeats on an ever-reduced scale. Fractal analysis is a useful method for their study. However, real-world fractals are not homogeneous; there is rarely an identical motif repeated on all scales. Their singularity may vary on different subsets; implying that these objects are multifractal. Multifractal analysis is a useful way to systematically characterize the spatial heterogeneity of both theoretical and experimental fractal patterns. However, the tools for multifractal analysis of objects in Euclidean space are not suitable for complex networks. In this thesis, we propose a new box covering algorithm for multifractal analysis of complex networks. This algorithm is demonstrated in the computation of the generalized fractal dimensions of some theoretical networks, namely scale-free networks, small-world networks, random networks, and a kind of real networks, namely PPI networks of different species. Our main finding is the existence of multifractality in scale-free networks and PPI networks, while the multifractal behaviour is not confirmed for small-world networks and random networks. As another application, we generate gene interactions networks for patients and healthy people using the correlation coefficients between microarrays of different genes. Our results confirm the existence of multifractality in gene interactions networks. This multifractal analysis then provides a potentially useful tool for gene clustering and identification. The third part of the thesis aims to investigate the topological properties of networks constructed from time series. Characterizing complicated dynamics from time series is a fundamental problem of continuing interest in a wide variety of fields. Recent works indicate that complex network theory can be a powerful tool to analyse time series. Many existing methods for transforming time series into complex networks share a common feature: they define the connectivity of a complex network by the mutual proximity of different parts (e.g., individual states, state vectors, or cycles) of a single trajectory. In this thesis, we propose a new method to construct networks of time series: we define nodes by vectors of a certain length in the time series, and weight of edges between any two nodes by the Euclidean distance between the corresponding two vectors. We apply this method to build networks for fractional Brownian motions, whose long-range dependence is characterised by their Hurst exponent. We verify the validity of this method by showing that time series with stronger correlation, hence larger Hurst exponent, tend to have smaller fractal dimension, hence smoother sample paths. We then construct networks via the technique of horizontal visibility graph (HVG), which has been widely used recently. We confirm a known linear relationship between the Hurst exponent of fractional Brownian motion and the fractal dimension of the corresponding HVG network. In the first application, we apply our newly developed box-covering algorithm to calculate the generalized fractal dimensions of the HVG networks of fractional Brownian motions as well as those for binomial cascades and five bacterial genomes. The results confirm the monoscaling of fractional Brownian motion and the multifractality of the rest. As an additional application, we discuss the resilience of networks constructed from time series via two different approaches: visibility graph and horizontal visibility graph. Our finding is that the degree distribution of VG networks of fractional Brownian motions is scale-free (i.e., having a power law) meaning that one needs to destroy a large percentage of nodes before the network collapses into isolated parts; while for HVG networks of fractional Brownian motions, the degree distribution has exponential tails, implying that HVG networks would not survive the same kind of attack.

On the robustness and security of digital image watermarking

In most of the digital image watermarking schemes, it becomes a common practice to address security in terms of robustness, which is basically a norm in cryptography. Such consideration in developing and evaluation of a watermarking scheme may severely affect the performance and render the scheme ultimately unusable. This paper provides an explicit theoretical analysis towards watermarking security and robustness in figuring out the exact problem status from the literature. With the necessary hypotheses and analyses from technical perspective, we demonstrate the fundamental realization of the problem. Finally, some necessary recommendations are made for complete assessment of watermarking security and robustness.

Next generation low-cost motion capture systems can provide comparable spatial accuracy to high-end systems

The objective quantification of three-dimensional kinematics during different functional and occupational tasks is now more in demand than ever. The introduction of new generation of low-cost passive motion capture systems from a number of manufacturers has made this technology accessible for teaching, clinical practice and in small/medium industry. Despite the attractive nature of these systems, their accuracy remains unproved in independent tests. We assessed static linear accuracy, dynamic linear accuracy and compared gait kinematics from a Vicon MX20 system to a Natural Point OptiTrack system. In all experiments data were sampled simultaneously. We identified both systems perform excellently in linear accuracy tests with absolute errors not exceeding 1%. In gait data there was again strong agreement between the two systems in sagittal and coronal plane kinematics. Transverse plane kinematics differed by up to 3 at the knee and hip, which we attributed to the impact of soft tissue artifact accelerations on the data. We suggest that low-cost systems are comparably accurate to their high-end competitors and offer a platform with accuracy acceptable in research for laboratories with a limited budget.

Probabilistic matching of image sets for video-based face recognition

We address the problem of face recognition on video by employing the recently proposed probabilistic linear discrimi-nant analysis (PLDA). The PLDA has been shown to be robust against pose and expression in image-based face recognition. In this research, the method is extended and applied to video where image set to image set matching is performed. We investigate two approaches of computing similarities between image sets using the PLDA: the closest pair approach and the holistic sets approach. To better model face appearances in video, we also propose the heteroscedastic version of the PLDA which learns the within-class covariance of each individual separately. Our experi-ments on the VidTIMIT and Honda datasets show that the combination of the heteroscedastic PLDA and the closest pair approach achieves the best performance.

Monte Carlo simulations of dynamic radiotherapy treatments

The effects of tumour motion during radiation therapy delivery have been widely investigated. Motion effects have become increasingly important with the introduction of dynamic radiotherapy delivery modalities such as enhanced dynamic wedges (EDWs) and intensity modulated radiation therapy (IMRT) where a dynamically collimated radiation beam is delivered to the moving target, resulting in dose blurring and interplay effects which are a consequence of the combined tumor and beam motion. Prior to this work, reported studies on the EDW based interplay effects have been restricted to the use of experimental methods for assessing single-field non-fractionated treatments. In this work, the interplay effects have been investigated for EDW treatments. Single and multiple field treatments have been studied using experimental and Monte Carlo (MC) methods. Initially this work experimentally studies interplay effects for single-field non-fractionated EDW treatments, using radiation dosimetry systems placed on a sinusoidaly moving platform. A number of wedge angles (60º, 45º and 15º), field sizes (20 × 20, 10 × 10 and 5 × 5 cm2), amplitudes (10-40 mm in step of 10 mm) and periods (2 s, 3 s, 4.5 s and 6 s) of tumor motion are analysed (using gamma analysis) for parallel and perpendicular motions (where the tumor and jaw motions are either parallel or perpendicular to each other). For parallel motion it was found that both the amplitude and period of tumor motion affect the interplay, this becomes more prominent where the collimator tumor speeds become identical. For perpendicular motion the amplitude of tumor motion is the dominant factor where as varying the period of tumor motion has no observable effect on the dose distribution. The wedge angle results suggest that the use of a large wedge angle generates greater dose variation for both parallel and perpendicular motions. The use of small field size with a large tumor motion results in the loss of wedged dose distribution for both parallel and perpendicular motion. From these single field measurements a motion amplitude and period have been identified which show the poorest agreement between the target motion and dynamic delivery and these are used as the „worst case motion parameters.. The experimental work is then extended to multiple-field fractionated treatments. Here a number of pre-existing, multiple–field, wedged lung plans are delivered to the radiation dosimetry systems, employing the worst case motion parameters. Moreover a four field EDW lung plan (using a 4D CT data set) is delivered to the IMRT quality control phantom with dummy tumor insert over four fractions using the worst case parameters i.e. 40 mm amplitude and 6 s period values. The analysis of the film doses using gamma analysis at 3%-3mm indicate the non averaging of the interplay effects for this particular study with a gamma pass rate of 49%. To enable Monte Carlo modelling of the problem, the DYNJAWS component module (CM) of the BEAMnrc user code is validated and automated. DYNJAWS has been recently introduced to model the dynamic wedges. DYNJAWS is therefore commissioned for 6 MV and 10 MV photon energies. It is shown that this CM can accurately model the EDWs for a number of wedge angles and field sizes. The dynamic and step and shoot modes of the CM are compared for their accuracy in modelling the EDW. It is shown that dynamic mode is more accurate. An automation of the DYNJAWS specific input file has been carried out. This file specifies the probability of selection of a subfield and the respective jaw coordinates. This automation simplifies the generation of the BEAMnrc input files for DYNJAWS. The DYNJAWS commissioned model is then used to study multiple field EDW treatments using MC methods. The 4D CT data of an IMRT phantom with the dummy tumor is used to produce a set of Monte Carlo simulation phantoms, onto which the delivery of single field and multiple field EDW treatments is simulated. A number of static and motion multiple field EDW plans have been simulated. The comparison of dose volume histograms (DVHs) and gamma volume histograms (GVHs) for four field EDW treatments (where the collimator and patient motion is in the same direction) using small (15º) and large wedge angles (60º) indicates a greater mismatch between the static and motion cases for the large wedge angle. Finally, to use gel dosimetry as a validation tool, a new technique called the „zero-scan method. is developed for reading the gel dosimeters with x-ray computed tomography (CT). It has been shown that multiple scans of a gel dosimeter (in this case 360 scans) can be used to reconstruct a zero scan image. This zero scan image has a similar precision to an image obtained by averaging the CT images, without the additional dose delivered by the CT scans. In this investigation the interplay effects have been studied for single and multiple field fractionated EDW treatments using experimental and Monte Carlo methods. For using the Monte Carlo methods the DYNJAWS component module of the BEAMnrc code has been validated and automated and further used to study the interplay for multiple field EDW treatments. Zero-scan method, a new gel dosimetry readout technique has been developed for reading the gel images using x-ray CT without losing the precision and accuracy.

Fast fourier analysis of structural organization in decellularized cartilage-on-bone laminates

Denaturation of tissues can provide a unique biological environment for regenerative medicine application only if minimal disruption of their microarchitecture is achieved during the decellularization process. The goal is to keep the structural integrity of such a construct as functional as the tissues from which they were derived. In this work, cartilage-on-bone laminates were decellularized through enzymatic, non-ionic and ionic protocols. This work investigated the effects of decellularization process on the microarchitecture of cartiligous extracellular matrix; determining the extent of how each process deteriorated the structural organization of the network. High resolution microscopy was used to capture cross-sectional images of samples prior to and after treatment. The variation of the microarchitecture was then analysed using a well defined fast Fourier image processing algorithm. Statistical analysis of the results revealed how significant the alternations among aforementioned protocols were (p < 0.05). Ranking the treatments by their effectiveness in disrupting the ECM integrity, they were ordered as: Trypsin> SDS> Triton X-100.

Anisotropy of articular cartilage reflects the ECM gradient architecture : Hough-Radon Transform Analysis

Tissue-specific extracellular matrix (ECM) is known to be an ideal bioscaffold to inspire the future of regenerative medicine. It holds the secret of how nature has developed such an organization of molecules into a unique functional complexity. This work exploited an innovative image processing algorithm and high resolution microscopy associated with mechanical analysis to establish a correlation between the gradient organization of cartiligous ECM and its anisotropic biomechanical response. This was hypothesized to be a reliable determinant that can elucidate how microarchitecture interrelates with biomechanical properties. Hough-Radon transform of the ECM cross-section images revealed its conformational variation from tangential interface down to subchondral region. As the orientation varied layer by layer, the anisotropic mechanical response deviated relatively. Although, results were in good agreement (Kendall's tau-b > 90%), there were evidences proposing that alignment of the fibrous network, specifically in middle zone, is not as random as it was previously thought.

Integrated vision-based attitude estimation

For many years, computer vision has lured researchers with promises of a low-cost, passive, lightweight and information-rich sensor suitable for navigation purposes. The prime difficulty in vision-based navigation is that the navigation solution will continually drift with time unless external information is available, whether it be cues from the appearance of the scene, a map of features (whether built online or known a priori), or from an externally-referenced sensor. It is not merely position that is of interest in the navigation problem. Attitude (i.e. the angular orientation of a body with respect to a reference frame) is integral to a visionbased navigation solution and is often of interest in its own right (e.g. flight control). This thesis examines vision-based attitude estimation in an aerospace environment, and two methods are proposed for constraining drift in the attitude solution; one through a novel integration of optical flow and the detection of the sky horizon, and the other through a loosely-coupled integration of Visual Odometry and GPS position measurements. In the first method, roll angle, pitch angle and the three aircraft body rates are recovered though a novel method of tracking the horizon over time and integrating the horizonderived attitude information with optical flow. An image processing front-end is used to select several candidate lines in a image that may or may not correspond to the true horizon, and the optical flow is calculated for each candidate line. Using an Extended Kalman Filter (EKF), the previously estimated aircraft state is propagated using a motion model and a candidate horizon line is associated using a statistical test based on the optical flow measurements and location of the horizon in the image. Once associated, the selected horizon line, along with the associated optical flow, is used as a measurement to the EKF. To evaluate the accuracy of the algorithm, two flights were conducted, one using a highly dynamic Uninhabited Airborne Vehicle (UAV) in clear flight conditions and the other in a human-piloted Cessna 172 in conditions where the horizon was partially obscured by terrain, haze and smoke. The UAV flight resulted in pitch and roll error standard deviations of 0.42° and 0.71° respectively when compared with a truth attitude source. The Cessna 172 flight resulted in pitch and roll error standard deviations of 1.79° and 1.75° respectively. In the second method for estimating attitude, a novel integrated GPS/Visual Odometry (GPS/VO) navigation filter is proposed, using a structure similar to a classic looselycoupled GPS/INS error-state navigation filter. Under such an arrangement, the error dynamics of the system are derived and a Kalman Filter is developed for estimating the errors in position and attitude. Through similar analysis to the GPS/INS problem, it is shown that the proposed filter is capable of recovering the complete attitude (i.e. pitch, roll and yaw) of the platform when subjected to acceleration not parallel to velocity for both the monocular and stereo variants of the filter. Furthermore, it is shown that under general straight line motion (e.g. constant velocity), only the component of attitude in the direction of motion is unobservable. Numerical simulations are performed to demonstrate the observability properties of the GPS/VO filter in both the monocular and stereo camera configurations. Furthermore, the proposed filter is tested on imagery collected using a Cessna 172 to demonstrate the observability properties on real-world data. The proposed GPS/VO filter does not require additional restrictions or assumptions such as platform-specific dynamics, map-matching, feature-tracking, visual loop-closing, gravity vector or additional sensors such as an IMU or magnetic compass. Since no platformspecific dynamics are required, the proposed filter is not limited to the aerospace domain and has the potential to be deployed in other platforms such as ground robots or mobile phones.