Artificial neural networks are systems composed of interconnected simple computing units known as artificial neurons which simulate some properties of their biological counterparts. They have been developed and studied for understanding how brains function, and for computational purposes. In order to use a neural network for computation, the network has to be designed in such a way that it performs a useful function. Currently, the most popular method of designing a network to perform a function is to adjust the parameters of a specified network until the network approximates the input-output behaviour of the function. Although some analytical knowledge about the function is sometimes available or obtainable, it is usually not used. Some neural network paradigms exist where such knowledge is utilized; however, there is no systematical method to do so. The objective of this research is to develop such a method. A systematic method of neural network design, which we call algebraic derivation methodology, is proposed and developed in this thesis. It is developed with an emphasis on designing neural networks to implement image processing algorithms. A key feature of this methodology is that neurons and neural networks are represented symbolically such that a network can be algebraically derived from a given function and the resulting network can be simplified. By simplification we mean finding an equivalent network (i.e., performing the same function) with fewer layers and fewer neurons. A type of neural networks, which we call LQT networks, are chosen for implementing image processing algorithms. Theorems for simplifying such networks are developed. Procedures for deriving such networks to realize both single-input and multiple-input functions are given. To show the merits of the algebraic derivation methodology, LQT networks for implementing some well-known algorithms in image processing and some other areas are developed by using the above mentioned theorems and procedures. Most of these networks are the first known such neural network models; in the case there are other known network models, our networks have the same or better performance in terms of computation time. / Applied Science, Faculty of / Electrical and Computer Engineering, Department of / Graduate

A model, which assumes tonal decomposition of textures, is proposed for texture analysis and segmentation. The model is based on an interpretation in which the tones form an orthogonal basis in a vector space of possible textures. Three methods for texture segmentation, which employ a texture indicator developed from this concept, are demonstrated. Application of the vector model requires the use of a priori information of the tonal frequencies present and the tonal amplitude distributions. A proposed algorithm, which uses only a priori knowledge of the spatial frequencies to extract the tonal amplitudes i.e. texture vectors, is discussed. Examp1es illustrating the application of this concept are presented. Finally a preliminary discussion of the problems associated with determining the tonal frequencies is presented / Master of Science

In this dissertation, we propose three methods to solve the problems of interactive image segmentation, video completion, and image denoising, which are all formulated as MRF-based energy minimization problems. In our algorithms, different MRF-based energy functions with particular techniques according to the characteristics of different tasks are designed to well fit the problems. With the energy functions, different optimization schemes are proposed to find the optimal results in these applications. In interactive image segmentation, an iterative optimization based framework is proposed, where in each iteration an MRF-based energy function incorporating an estimated initial probabilistic map of the image is optimized with a relaxed global optimal solution. In video completion, a well-defined MRF energy function involving both spatial and temporal coherence relationship is constructed based on the local motions calculated in the first step of the algorithm. A hierarchical belief propagation optimization scheme is proposed to efficiently solve the problem. In image denoising, label relaxation based optimization on a Gaussian MRF energy is used to achieve the global optimal closed form solution. / Many problems in computer vision involve assigning each pixel a label, which represents some spatially varying quantity such as image intensity in image denoising or object index label in image segmentation. In general, such quantities in image processing tend to be spatially piecewise smooth, since they vary smoothly in the object surface and change dramatically at object boundaries, while in video processing, additional temporal smoothness is satisfied as the corresponding pixels in different frames should have similar labels. Markov random field (MRF) models provide a robust and unified framework for many image and video applications. The framework can be elegantly expressed as an MRF-based energy minimization problem, where two penalty terms are defined with different forms. Many approaches have been proposed to solve the MRF-based energy optimization problem, such as simulated annealing, iterated conditional modes, graph cuts, and belief propagation. / Promising results obtained by the proposed algorithms, with both quantitative and qualitative comparisons to the state-of-the-art methods, demonstrate the effectiveness of our algorithms in these image and video processing applications. / Liu, Ming. / Adviser: Xiaoou Tang. / Source: Dissertation Abstracts International, Volume: 72-04, Section: B, page: . / Thesis (Ph.D.)--Chinese University of Hong Kong, 2010. / Includes bibliographical references (leaves 79-89). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Electronic reproduction. Ann Arbor, MI : ProQuest Information and Learning Company, [200-] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Electronic reproduction. Ann Arbor, MI : ProQuest Information and Learning Company, [200-] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Abstract also in Chinese.

In the brain mapping case, the geodesic closest points are used as the anatomical constraints for the inter-subject non-rigid registration. The method uses the anatomical constraint in the non-rigid registration which is much more reasonable for the anatomical correspondence. The registration result shows significant improvement comparing with the iterative closest points based method. / In the third application, we use the non-rigid registration method to register the different sweeps of freehand ultrasound images. We setup a 3D freehand ultrasound imaging system to capture images of a human anatomy such as liver, prostate, brain tumor and fetus. The arbitrary scanned image slices are reconstructed and resliced into volumetric dataset. We use a B-spline based non-rigid registration method to compounding different freehand ultrasound sweeps. This technique can be used to make 3D ultrasound models of fetus and other organs. / Medical image registration is an active research area during the last two decades. The registration technique can be widely used in the applications of the computer aided surgery, brain mapping and pathological detection and analysis. With the development of the computing power, fast and accurate registration techniques have been developed into necessary tools for quantitative analysis of the medical image. / Non-rigid registration methods can be used in atlas based image segmentation, inter-subject brain image registration and 3D freehand ultrasound modeling. In one of our proposed novel segmentation methods, we interleave the segmentation and the registration processes by using the segmentation to provide the anatomical constraints for registration to improve the atlas based non-rigid registration. This updated registration can be used to improve the new segmentation. This process is repeated until a good result in segmentation is obtained. / The registration methods can be classified into rigid and non-rigid registrations according to whether the anatomy is locally deformed or not. According to the sensor by which the images are taken, the registration will be divided into mono-modal and multi-modal image registration. Since the invention of the medical imaging devices, great diversity of medical imaging sensors have been developed with different physical principles. In practice we have to face the problem of multi-modal registration. In medical image analysis, we often have to consider the images of the human anatomy with deformable characteristics. In order to achieve this goal we need to use the voxel based registration method which considers all of the voxel information of the images in matching. There are several non-rigid registration approaches. However, the variational approach of non-rigid registration can represent the registration problem into a well-posed problem with a well-founded mathematical base. In our work, we considered the forward and backward deformation functions and proposed a variational approach for a new consistent multi-modal non-rigid registration method. By this way, we will find the forward and backward transform to be close to the inverse of each other. This makes the correspondence between two images more consistent and accurate. We use both explicit and implicit difference method to solve the Euler-Lagrange equation and the results show significant improvements in the transformation inverse consistency. Although variational approach for multi-modal non-rigid registration can solve the non-rigid registration problem well, generally speaking, it is slow. The displacement of each voxel has to be calculated and the iteration time is very long since the number of the unknowns are large. Although a multi-resolution strategy can be used to speed up, the registration problem is still slow when registering large medical datasets. The 3D B-spline based method has been used as an efficient method to register medical images since only a small number of control points are used to manipulate the local deformation field. In our work, we developed a 3D B-spline based consistent multi-modal non-rigid registration method with an explicit representation of derivatives. The conventional optimization methods can be used to find the optimal parameters. We use a hierarchical B-spline refinement method to approximate the deformation function from larger to smaller scale. Since the derivatives of the cost function is represented in an explicit way, the computing is reduced. It is more efficient than directly computing the derivative of the cost function by using a numerical evaluation method. The method can be considered as a multi-grid method for solving the consistent variational registration problem. The computing speed is increased by several times. The B-spline based method needs far less iterations to converge as its number of unknowns is small. / Zhang Zhijun. / "October 2005." / Source: Dissertation Abstracts International, Volume: 67-11, Section: B, page: 6645. / Thesis (Ph.D.)--Chinese University of Hong Kong, 2005. / Includes bibliographical references (p. 209-233). / Electronic reproduction. Hong Kong : Chinese University of Hong Kong, [2012] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Electronic reproduction. [Ann Arbor, MI] : ProQuest Information and Learning, [200-] System requirements: Adobe Acrobat Reader. Available via World Wide Web. / Abstracts in English and Chinese. / School code: 1307.

Jiang, Lei. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2011. / Includes bibliographical references (p. 68-75). / Abstracts in English and Chinese. / Chapter 1 --- Introduction --- p.1 / Chapter 2 --- Background and Motivation --- p.5 / Chapter 2.1 --- Foreshortening --- p.5 / Chapter 2.1.1 --- Vanishing Point --- p.6 / Chapter 2.1.2 --- Metric Rectification --- p.8 / Chapter 2.2 --- Content Aware Image Resizing --- p.11 / Chapter 2.3 --- Texture Deformation --- p.15 / Chapter 2.3.1 --- Shape from texture --- p.16 / Chapter 2.3.2 --- Shape from lattice --- p.18 / Chapter 3 --- Resizing on Facade --- p.21 / Chapter 3.1 --- Introduction --- p.21 / Chapter 3.2 --- Related Work --- p.23 / Chapter 3.3 --- Algorithm --- p.24 / Chapter 3.3.1 --- Facade Detection --- p.25 / Chapter 3.3.2 --- Facade Resizing --- p.32 / Chapter 3.4 --- Results --- p.34 / Chapter 4 --- Cell Texture Editing --- p.42 / Chapter 4.1 --- Introduction --- p.42 / Chapter 4.2 --- Related Work --- p.44 / Chapter 4.3 --- Our Approach --- p.46 / Chapter 4.3.1 --- Cell Detection --- p.47 / Chapter 4.3.2 --- Local Affine Estimation --- p.49 / Chapter 4.3.3 --- Affine Transformation Field --- p.52 / Chapter 4.4 --- Photo Editing Applications --- p.55 / Chapter 4.5 --- Discussion --- p.58 / Chapter 5 --- Conclusion --- p.65 / Bibliography --- p.67

Chen, Jianing. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2009. / Includes bibliographical references (leaves 47-54). / Abstract also in Chinese. / Chapter 1 --- Introduction --- p.1 / Chapter 1.1 --- Background --- p.1 / Chapter 1.2 --- Related Work --- p.3 / Chapter 1.3 --- Thesis Organization --- p.5 / Chapter 2 --- Review on Optical Flow Estimation --- p.6 / Chapter 2.1 --- Variational Model --- p.6 / Chapter 2.1.1 --- Basic Assumptions and Constraints --- p.6 / Chapter 2.1.2 --- More General Energy Functional --- p.9 / Chapter 2.2 --- Discontinuity Preserving Techniques --- p.9 / Chapter 2.2.1 --- Data Term Robustification --- p.10 / Chapter 2.2.2 --- Diffusion Based Regularization --- p.11 / Chapter 2.2.3 --- Segmentation --- p.15 / Chapter 2.3 --- Chapter Summary --- p.15 / Chapter 3 --- Segmentation Based Optical Flow Estimation --- p.17 / Chapter 3.1 --- Initial Flow --- p.17 / Chapter 3.2 --- Color-Motion Segmentation --- p.19 / Chapter 3.3 --- Parametric Flow Estimating Incorporating Segmentation --- p.21 / Chapter 3.4 --- Confidence Map Construction --- p.24 / Chapter 3.4.1 --- Occlusion detection --- p.24 / Chapter 3.4.2 --- Pixel-wise motion coherence --- p.24 / Chapter 3.4.3 --- Segment-wise model confidence --- p.26 / Chapter 3.5 --- Final Combined Variational Model --- p.28 / Chapter 3.6 --- Chapter Summary --- p.28 / Chapter 4 --- Experiment Results --- p.30 / Chapter 4.1 --- Quantitative Evaluation --- p.30 / Chapter 4.2 --- Warping Results --- p.34 / Chapter 4.3 --- Chapter Summary --- p.35 / Chapter 5 --- Application - Single Image Animation --- p.37 / Chapter 5.1 --- Introduction --- p.37 / Chapter 5.2 --- Approach --- p.38 / Chapter 5.2.1 --- Pre-Process Stage --- p.39 / Chapter 5.2.2 --- Coordinate Transform --- p.39 / Chapter 5.2.3 --- Motion Field Transfer --- p.41 / Chapter 5.2.4 --- Motion Editing and Apply --- p.41 / Chapter 5.2.5 --- Gradient-domain composition --- p.42 / Chapter 5.3 --- Experiments --- p.43 / Chapter 5.3.1 --- Active Motion Transfer --- p.43 / Chapter 5.3.2 --- Animate Stationary Temporal Dynamics --- p.44 / Chapter 5.4 --- Chapter Summary --- p.45 / Chapter 6 --- Conclusion --- p.46 / Bibliography --- p.47

No description available.

In this thesis, methods for achieving finer scale multi-spectral classification through the use of super resolution (SR) techniques are investigated. A new super resolution algorithm Maximum a Posteriori based on the universal hidden Markov tree model (MAP-uHMT) is developed which can be applied successfully to super-resolve each multi-spectral channel before classification by standard methods. It is believed that this is the first time that a true super resolution algorithm has been applied to multi-spectral classification, and results are shown to be excellent. Image registration is an important step for SR in which misalignment can be measured for each of many low resolution images; therefore, a new and computationally efficient image registration is developed for this particular application. This improved elastic image registration method can deal with a global affine warping and local shift translations based on coarse to fine pyramid levels. The experimental results show that it can provide good registration accuracy in less computational time than comparable methods. Maximum a posteriori (MAP) is adopted to deal with the ill-conditioned problem of super resolution, wherein a prior is constructed based on the universal hidden Markov tree (uHMT) model in the wavelet domain. In order to test this prior for MAP estimation, it is first tested in the simpler and typically ill-conditioned problem of image denoising. Experimental results illustrate that this new image denoising method achieves good performance for the test images. The new prior is then extended to SR. By combining with the new elastic image registration algorithm, MAP-uHMT can super resolve both some natural video frames and remote sensing images. Test results with both synthetic data and real data show that this method achieves super resolution both visually and quantitatively. In order to show that MAPuHMT is also applicable more widely, it is tested on a sequence of long-range surveillance images captured under conditions of atmospheric turbulence distortion. The results suggest that super resolution may have been achieved in this application also.

Thesis (Ph.D.)--Massachusetts Institute of Technology, Dept. of Electrical Engineering and Computer Science, 1982. / MICROFICHE COPY AVAILABLE IN ARCHIVES AND ENGINEERING. / Bibliography: leaves 336-341. / by David John Rossi. / Ph.D.

Yu, Tsz Ho. / Thesis (M.Phil.)--Chinese University of Hong Kong, 2009. / Includes bibliographical references (leaves 95-100). / Abstract also in Chinese. / Chapter 1 --- Introduction --- p.2 / Chapter 1.1 --- Surveillance and Computer Vision --- p.3 / Chapter 1.2 --- The Need for Abnormal Behavior Detection --- p.3 / Chapter 1.2.1 --- The Motivation --- p.3 / Chapter 1.2.2 --- Choosing the Right Surveillance Target --- p.5 / Chapter 1.3 --- Abnormal Behavior Detection: An Overview --- p.6 / Chapter 1.3.1 --- Challenges in Detecting Abnormal Behaviors --- p.6 / Chapter 1.3.2 --- Limitations of Existing Approaches --- p.8 / Chapter 1.3.3 --- New Design Concepts --- p.9 / Chapter 1.3.4 --- Requirements for Abnormal Behavior Detection --- p.10 / Chapter 1.4 --- Contributions --- p.11 / Chapter 1.4.1 --- An Unsupervised Experience-based Approach for Abnormal Behavior Detection --- p.11 / Chapter 1.4.2 --- Motion Histogram Transform: A Novel Feature Descriptors --- p.12 / Chapter 1.4.3 --- Real-time Algorithm for Abnormal Behavior Detection --- p.12 / Chapter 1.5 --- Thesis Organization --- p.13 / Chapter 2 --- Literature Review --- p.14 / Chapter 2.1 --- From Segmentation to Visual Tracking --- p.14 / Chapter 2.1.1 --- Environment Modeling and Segmentation --- p.15 / Chapter 2.1.2 --- Spatial-temporal Feature Extraction --- p.18 / Chapter 2.2 --- Detecting Irregularities in Videos --- p.21 / Chapter 2.2.1 --- Model-based Method --- p.22 / Chapter 2.2.2 --- Non Model-based Method --- p.26 / Chapter 3 --- Design Framework --- p.29 / Chapter 3.1 --- Dynamic Scene and Behavior Model --- p.30 / Chapter 3.1.1 --- Images Sequences and Video --- p.30 / Chapter 3.1.2 --- Motions and Behaviors in Video --- p.31 / Chapter 3.1.3 --- Discovering Abnormal Behavior --- p.32 / Chapter 3.1.4 --- Problem Definition --- p.33 / Chapter 3.1.5 --- System Assumption --- p.34 / Chapter 3.2 --- Methodology --- p.35 / Chapter 3.2.1 --- Potential Improvements --- p.35 / Chapter 3.2.2 --- The Design Framework --- p.36 / Chapter 4 --- Implementation --- p.40 / Chapter 4.1 --- Preprocessing --- p.40 / Chapter 4.1.1 --- Data Input --- p.41 / Chapter 4.1.2 --- Motion Detection --- p.41 / Chapter 4.1.3 --- The Gaussian Mixture Background Model --- p.43 / Chapter 4.2 --- Feature Extraction --- p.46 / Chapter 4.2.1 --- Optical Flow Estimation --- p.47 / Chapter 4.2.2 --- Motion Histogram Transforms --- p.53 / Chapter 4.3 --- Feedback Learning --- p.56 / Chapter 4.3.1 --- The Observation Matrix --- p.58 / Chapter 4.3.2 --- Eigenspace Transformation --- p.58 / Chapter 4.3.3 --- Self-adaptive Update Scheme --- p.61 / Chapter 4.3.4 --- Summary --- p.62 / Chapter 4.4 --- Classification --- p.63 / Chapter 4.4.1 --- Detecting Abnormal Behavior via Statistical Saliencies --- p.64 / Chapter 4.4.2 --- Determining Feedback --- p.65 / Chapter 4.5 --- Localization and Output --- p.66 / Chapter 4.6 --- Conclusion --- p.69 / Chapter 5 --- Experiments --- p.71 / Chapter 5.1 --- Experiment Setup --- p.72 / Chapter 5.2 --- A Summary of Experiments --- p.74 / Chapter 5.3 --- Experiment Results: Part 1 --- p.78 / Chapter 5.4 --- Experiment Results: Part 2 --- p.81 / Chapter 5.5 --- Experiment Results: Part 3 --- p.83 / Chapter 5.6 --- Experiment Results: Part 4 --- p.86 / Chapter 5.7 --- Analysis and Conclusion --- p.86 / Chapter 6 --- Conclusions --- p.88 / Chapter 6.1 --- Application Extensions --- p.88 / Chapter 6.2 --- Limitations --- p.89 / Chapter 6.2.1 --- Surveillance Range --- p.89 / Chapter 6.2.2 --- Preparation Time for the System --- p.89 / Chapter 6.2.3 --- Calibration of Background Model --- p.90 / Chapter 6.2.4 --- Instability of Optical Flow Feature Extraction --- p.91 / Chapter 6.2.5 --- Lack of 3D information --- p.91 / Chapter 6.2.6 --- Dealing with Complex Behavior Patterns --- p.92 / Chapter 6.2.7 --- Potential Improvements --- p.92 / Chapter 6.2.8 --- New Method for Classification --- p.93 / Chapter 6.2.9 --- Introduction of Dynamic Texture as a Feature --- p.93 / Chapter 6.2.10 --- Using Multiple-camera System --- p.93 / Chapter 6.3 --- Summary --- p.94 / Bibliography --- p.95