Browsing by Subject "Pattern recognition systems"
Now showing 1 - 19 of 19
Results Per Page
Sort Options
Item A modified Kanade model for three-dimensional object recognition(Texas Tech University, 1990-05) Lau, Henry Chi-ming--Item A study of statistical image classification and enhancement(Texas Tech University, 1984-08) Tzeng, Mien-hueiNot availableItem A VLSI optical detector array employing heterodyne detection(Texas Tech University, 1997-05) Soni, Tejvansh SinghThe integration of image sensors with circuitry for driving the image sensor and performing on-chip signal processing is becoming increasingly popular for a multitude of signal processing applications. A high degree of on-chip signal processing helps enable miniaturization of instrument systems and simplify system interfaces. In this work, the design of a powerful and versatile VLSI optical sensor array, with on-chip circuitry to perform temporal electronic heterodyne detection on a pixel-bypixel basis is presented. Heterodyne detection techniques significantly enhance the dynamic range and signal to noise ratio, as compared to base-band detectors. The unavailability of heterodyne detector arrays has been a bottleneck in many imageprocessing systems, restricting the use of heterodyne detection techniques to scanning based systems, or systems having a single temporal output, such as acousto-optic space integrating correlators and convolvers. The need for heterodyne detector arrays in acousto-optics has been emphasized by prominent researchers in the field.Item Adaptive clustering for image segmentation(Texas Tech University, 1998-12) Neeruganti, JagadeeshThe purpose of image segmentation is to separate different objects embedded in an image. Many image segmentation techniques are available in the literature. Some of the simple techniques employ thresholding based on the gray level histogram, while a number of other sophisticated techniques have been developed in recent years. Among the recent techniques, limited success has been achieved by employing some fuzzy selfsupervised neural networks for object extraction. This work reviews the basic segmentation techniques and demonstrates the applications of adaptive clustering techniques, which make use of neural networks and fuzzy methods for image segmentation. The adaptive clustering techniques used are two neuro-fuzzy techniques namely, the Integrated Adaptive Fuzzy Clustering (lAFC) and Adaptive Fuzzy Leader Clustering (AFLC). The performances of these techniques are compared with the performance of the fuzzy c-means (FCM algorithm as applied to image segmentation.Item Adaptive hierarchical classification with limited training data(2002) Morgan, Joseph Troy; Crawford, Melba M.This research focused on the development of a hierarchical approach for classification that is robust with respect to training data that are limited both in quantity and spatial extent. Many difficult classification problems involve a high dimensional input and output space (candidate labels). Due to the "curse of dimensionality," it is necessary to reduce the size of the input space when there is only a limited quantity of training data available. While a significant amount of research has focused on transforming the input space into a reduced feature space that accurately discriminates between the classes in a fixed output space, traditional approaches fail to capitalize on the domain knowledge and flexibility gained by transforming the feature space and the output space simultaneously. A new approach is proposed that utilizes domain knowledge, which is automatically discovered from the data, to combat the "small sample size" problem. Spatially limited training data can result in poor inference concerning the true populations. The detrimental impact that can result if this issue is ignored is explored and demonstrated. Transferal of information that was previously acquired is used to update the signatures with the new clusters if the hypothesis that the new clusters are indeed just deformed versions of what already exists in the spectral library is accepted. Independent of limited training data, both in terms of the spatial implications and limited quantity, different sampling subsets of the same ground truth may result in slightly different classifiers. This issue has not been addressed rigorously. The advantages gained by using an ensemble of classifiers built from sub-samples of training data are widely acknowledged but have not previously been used in the context of a hierarchical classifier for remote sensing data or for hyperspectral data in general. The ensemble of classifiers is used to identify a suitable level of the tree for situations where the resolution of the output space cannot be supported. Further decisions of how the classification structure should be adapted and at what level need to be made are explored. Furthermore, pseudolabeled data are utilized to improve classification results at that level of resolution.Item An integrated adaptive fuzzy clustering model for pattern recognition(Texas Tech University, 1993-12) Kim, Yong SooThis dissertation presents the Integrated Adaptive Fuzzy Clustering (lAFC) model that overcomes the abovementioned problems. The lAFC model is a fuzzy neural network which incorporates a new fuzzy learning rule into a neural network structure, similar to the ART-1 neural network. The new learning rule incorporates fusion of a fuzzy membership value, the 7C-function [35] , and a function of the number of iterations into the incremental learning rule. The combination of the 7i-function and a function of the number of iterations guarantees weights to converge. The lAFC model introduces a new similarity measure that incorporates a fuzzy membership value into the Euclidean distance. The Euclidean distance and the Mahalanobis distance are commonly used as similarity measures. Even though the use of the Euclidean distance is convenient, it is best suited to hyper-spherical cluster shapes. The Mahalanobis distance accounts for some variations in cluster shape, but it works well for only hyper-ellipsoidal cluster shapes and is computationally burdensome. The new similarity measure considers not only the distance between the input data point and the centroid of a winning cluster but also the relative location of the input point to the existing cluster centroids as the degree of similarity. Thus, it gives more flexibility to the shape of clusters formed. Chapter II describes current clustering and fuzzy clustering algorithms. Problems of current algorithms are also discussed. Chapter III provides an overview of current self-organizing neural networks for clustering. Chapter IV discusses current neuro-fuzzy clustering algorithms. Chapter V describes the new similarity measure and the lAFC model. Chapter VI presents the results of classification of real data sets by the lAFC model and compares the performance of the lAFC model with that of other recent neuro-fuzzy clustering and fuzzy clustering algorithms. Chapter VII concludes the dissertation.Item Automated object recognition through reinforcement learning(Texas Tech University, 2002) Ge, Zhanyu; Mitra, Sunanda; Krile, Thomas; Bredeson, Jon G.Object recognition, a branch of pattern recognition, is to identify and localize one or more objects in a given scene. We have to determine what is present and where it is within the input image. Although great achievements have been made during the last decades, currently existing object recognition techniques have shortcomings like unreliability and inefficiency, general inadaptability, manual template marking heavily influenced by human factors, inability to recognize an object without a model, and so on. Any recognition problem can be formulated as a searching process and has to be guided in a controlled manner. All search problems involve optimization, so object recognition requires optimization and control techniques. Reinforcement learning is learning how to behave given a situation and possible actions to maximize the total expected reward in the long run, and therefore needs to be optimized. Most pattern recognition techniques do not combine reinforcement learning for feature understanding. In this dissertation, reinforcement learning is applied both to automatic template generation from a model image and to template matching within the input image. The newly designed affine parameter estimation algorithm provides reliable results based on information contained at all feature point locations. The points are extracted in the scale-space using isophote curvature extreme points, which are invariant to affine transformations. The affine parameter estimation algorithm is applicable to any kind of translations, rotations, and scales, and moderate occlusions and deformations of the object to be recognized. Experiment results showed that the proposed set of algorithms are fast, efficient, and potentially robust. The automatic template generation algorithm, an efficient contour tracing one in gray-level images, can also be used in object recognition without a model. This is a new research field, and a great amount of future work needs to be done before an intelligent recognition system, as efficient as the human vision system, can be developed.Item Calibration and three-dimensional reconstruction using epipolar constraints on a structured light computer vision system(Texas Tech University, 1997-05) Lin, ChangxingA new structured light computer vision system was developed to determine 3 dimensional geometry information of objects. The system was composed of a dot matrix pattern laser projector, and two cameras (labeled as A and B). Here, the camera A is called main camera. The cameras (B) functions as a checking device to determine the correct image matching between the main image and the projector, so it is called checking camera. There are three contributions in this dissertation and they are as follows: First, a new camera calibration technique is provided, in which the image center, uncertainty scale factor, camera focal length, rotation matrix, and translation vector can be determined using at least seven noncoplanar calibration points; the orthogonality of rotation matrix can be satisfied not only theoretical but also numerically with actual calibration; all intrinsic and extrinsic parameters can be determined using the same set of data; no assumption is needed for the world coordinate system setup; and no nonlinear techniques are required. Second, a new linear approach for estimating the epipolar lines on the main camera (A), related to the projector, is developed. The existing methods can not guarantee that all image points on the same epipolar line on the main camera, related to the projector, have the same corresponding epipolar line on the projector. This is against the epipolar geometric constraints. The approach developed here can guarantee that all points on the same epipolar line on the main image, related to the projector, must have the same corresponding epipolar line on the projector. Third, two checking point equations are given to determine the correct image matching among the main image, the checking image, and the projector. The methods developed here, only require use of the epipolar lines on the projector, related to the main camera (A). Calibration of the projector is not required. A review of the state of the art is given in the first three chapters. All methods developed here were verified experimentally.Item Characterization of digital phase-locked loops(Texas Tech University, 2003-05) Vepa, Sri Kiran V SPhase-locked loops are a relatively new class of circuits used primarily in communication applications. The capture range of a phase-locked loop is a critical parameter because it trades directly with the loop bandwidth. Different architectures for the phase-locked loop (PLL) have been proposed which can broaden the capture range (1-3]. However, in most of the research, very little emphasis was made on studying the exact dependence of the capture range on the different circuit parameters, which define the individual components of a phase-locked loop. The effect of these parameters, for instance, the W/L ratio of the transistors, can be prominent. This thesis is aimed at designing a circuit for a digital phase locked loop, characterizing the components and discussing a method of estimating the capture range. This circuit can act as a starting point in solving the above mentioned problem. The next step would be to observe the dependence of capture range on circuit parameters.Item Feature-based exploitation of multidimensional radar signatures(2008-08) Raynal, Ann Marie; Ling, HaoAn important problem in electromagnetics is that of extracting, interpreting, and exploiting scattering mechanisms from the scattered field of a target. Termed “features”, these physics-based descriptions of scattering phenomenology have many and diverse applications such as target identification, classification, validation, and imaging. In this dissertation, the feature extraction, analysis, and exploitation of both synthetic and measured multidimensional radar signatures are investigated. Feature extraction is first performed on simulated data of the highfrequency electromagnetics solver Xpatch. The scattered, far-field of an electrically large target is well-approximated by a discrete set of points known as scattering centers. Xpatch yields three-dimensional (3D) scattering centers of a target one aspect angle at a time by using the shooting and bouncing ray technique and a computer-aided design (CAD) model of the target. The feature extraction technique groups scattering centers across multiple angles that pertain to the same scattering mechanism. Using a nearest neighbor clustering algorithm, this association is carried-out in a multidimensional grid of scattering center angle, bounce, and spatial location, wherein distinct scattering mechanisms are assumed to be non-overlapping. Synthetic monostatic and bistatic feature sets are extracted and analyzed using this algorithm. Additionally, feature sets are exploited to assist humans in electromagnetic CAD model validation. The generation of target CAD models is a challenging, resource-limited, and human-experience-based process. Target features extracted from a CAD model in question are compared individually to measured data from the physical target by projection of their radar signatures. CAD model disagreements such as missing, added, or dimensionally inaccurate components, as well as measurement imperfections are analyzed. Target traceback information of the features identifies flawed areas of the model. The projection value quantifies the degree of disagreement. The feature extraction methodology is next modified for measured radar signatures which lack readily available scattering center and bounce information. First, many ground plane synthetic aperture radar images of overlapping, limited apertures in azimuth are formed from the measurement data. Then, two-dimensional scattering centers of all images are estimated using a modified CLEAN algorithm. Feature extraction is lastly performed as with Xpatch data, though a reduction in grid dimensionality and orthogonality occurs. Finally, measured feature sets are exploited for sparse elevation 3D imaging and improved CAD model validation. The first application estimates the truth 3D scattering center of each feature using linear least squares to then visualize a composite 3D image of the target. The second application projects both synthetic and measured feature radar signatures to mitigate errors from the intersection of features in the combined measurement signature.Item Foveated object recognition by corner search(2008-05) Arnow, Thomas Louis, 1946-; Bovik, Alan C. (Alan Conrad), 1958-; Geisler, Wilson S.Here we describe a gray scale object recognition system based on foveated corner finding, the computation of sequential fixation points, and elements of Lowe’s SIFT transform. The system achieves rotational, transformational, and limited scale invariant object recognition that produces recognition decisions using data extracted from sequential fixation points. It is broken into two logical steps. The first is to develop principles of foveated visual search and automated fixation selection to accomplish corner search. The result is a new algorithm for finding corners which is also a corner-based algorithm for aiming computed foveated visual fixations. In the algorithm, long saccades move the fovea to previously unexplored areas of the image, while short saccades improve the accuracy of putative corner locations. The system is tested on two natural scenes. As an interesting comparison study we compare fixations generated by the algorithm with those of subjects viewing the same images, whose eye movements are being recorded by an eyetracker. The comparison of fixation patterns is made using an information-theoretic measure. Results show that the algorithm is a good locator of corners, but does not correlate particularly well with human visual fixations. The second step is to use the corners located, which meet certain goodness criteria, as keypoints in a modified version of the SIFT algorithm. Two scales are implemented. This implementation creates a database of SIFT features of known objects. To recognize an unknown object, a corner is located and a feature vector created. The feature vector is compared with those in the database of known objects. The process is continued for each corner in the unknown object until enough information has been accumulated to reach a decision. The system was tested on 78 gray scale objects, hand tools and airplanes, and shown to perform well.Item Investigating the use of tabu search to find near-optimal solutions in multiclassifier systems(2003) Korycinski, Donna Kay; Crawford, Melba M.; Barnes, J. Wesley.Item Learnable similarity functions and their application to record linkage and clustering(2006) Bilenko, Mikhail Yuryevich; Mooney, Raymond J. (Raymond Joseph)Many machine learning and data mining tasks depend on functions that estimate similarity between instances. Similarity computations are particularly important in clustering and information integration applications, where pairwise distances play a central role in many algorithms. Typically, algorithms for these tasks rely on pre-defined similarity measures, such as edit distance or cosine similarity for strings, or Euclidean distance for vector-space data. However, standard distance functions are frequently suboptimal as they do not capture the appropriate notion of similarity for a particular domain, dataset, or application. In this thesis, we present several approaches for addressing this problem by employing learnable similarity functions. Given supervision in the form of similar or disviii similar pairs of instances, learnable similarity functions can be trained to provide accurate estimates for the domain and task at hand. We study the problem of adapting similarity functions in the context of several tasks: record linkage, clustering, and blocking. For each of these tasks, we present learnable similarity functions and training algorithms that lead to improved performance. In record linkage, also known as duplicate detection and entity matching, the goal is to identify database records referring to the same underlying entity. This requires estimating similarity between corresponding field values of records, as well as overall similarity between records. For computing field-level similarity between strings, we describe two learnable variants of edit distance that lead to improvements in linkage accuracy. For learning record-level similarity functions, we employ Support Vector Machines to combine similarities of individual record fields in proportion to their relative importance, yielding a high-accuracy linkage system. We also investigate strategies for efficient collection of training data which can be scarce due to the pairwise nature of the record linkage task. In clustering, similarity functions are essential as they determine the grouping of instances that is the goal of clustering. We describe a framework for integrating learnable similarity functions within a probabilistic model for semi-supervised clustering based on Hidden Markov Random Fields (HMRFs). The framework accommodates learning various distance measures, including those based on Bregman divergences (e.g., parameterized Mahalanobis distance and parameterized KL-divergence), as well as directional measures (e.g., cosine similarity). Thus, it is applicable to a wide range of domains and data representations. Similarity functions are learned within the HMRF-KMEANS algorithm derived from the framework, leading to significant improvements in clustering accuracy. The third application we consider, blocking, is critical in making record linkage and clustering algorithms scalable to large datasets, as it facilitates efficient selection of approximately similar instance pairs without explicitly considering all possible pairs. Previously proposed blocking methods require manually constructing a similarity function or a set of similarity predicates, followed by hand-tuning of parameters. We propose learning blocking functions automatically from linkage and semi-supervised clustering supervision, which allows automatic construction of blocking methods that are efficient and accurate. This approach yields computationally cheap learnable similarity functions that can be used for scaling up in a variety of tasks that rely on pairwise distance computations, including record linkage and clustering.Item Morphological filters for image enhancement(Texas Tech University, 1996-12) Kher, AlokDigital images are subjected to filtering processes during the operations of noise reduction and lossy compression. Fine details are often lost or severely altered in these filtering processes. Connectivity preserving morphological filters haven been proposed in the past to remove noise while preserving thin but connected regions However, these filters preserved regional connectivity only in restricted orientations The present work has developed morphological filters that may be used for fast and efficient removal of noise while completely preserving connectivity information in gray scale images. These filters are shown to satisfy the requirements of well behaved abstract operations of algebraic opening and closing. When applied to the problem of speckle noise reduction from synthetic aperture radar images, the new filters performed significantly better than conventional linear and non-linear filters. The present work has also developed an image representation approach that may be used for developing high quality lossy image compression techniques based on morphological muhiresolution pyramid decomposition of images. A pyramid decomposition technique represents an image as a pyramid of differential images which store incremental information at various resolutions The lossy compression techniques based on pyramid decomposition often discard the first differential image component which usually consists of a substantial amount of high frequency noise. Complete omission of this image component can result in the loss of fine image details. The present work has developed an approach to approximately reconstruct the first differential image from its two components consisting of directional information. The simplification process is shown to be equivalent to connectivity preserving filtering. For various standard images, the entropies of the differential images were shown to decrease by 35% to 40% for approximately 10% mean square error between the original and the reconstructed differential images.Item Neural network structure modeling: an application to font recognition(Texas Tech University, 1988-12) Lee, Ming-chih YehTwo neural network models, Model H-Hl (Hogg and Huberman, 1984) and Model H-H2 (Hogg and Huberman, 1985) have been successfully applied to the font recognition problem and were used to recognize 26 English capital letters, each with six font representations. Recognition rate, memory space requirement, learning speed, and recognition speed were used to measure the models* performances. Model parameters such as memory array size, Smin_Smax, and Mmin_Mmax were varied to elucidate the models' behavior. As a result, both models achieved a 100% recognition rate when all six fonts were used as the training as well as the recognition set. When three out of six fonts were used for training, Model H-Hl achieved a maximum recognition rate of 87.82% and Model H-H2 achieved a maximum recognition rate of 89.10%. This shows that the basins of attractor states existed for the letters in most of the various font presentations. Model H-H2 significantly outperformed Model H-Hl in terms of recognition rate, use of memory space, and learning speed when all six fonts were used as the training set. This was supported by the results of the Pairwised T Test.Item Optimal visual search strategies using natural scene statistics(2007-12) Raj, Raghu G., 1975-; Bovik, Alan C. (Alan Conrad), 1958-I present theoretical foundations and perform computational studies on optimal search strategies in natural scenes performed by foveated artificial vision systems, based on novel characterizations of Natural Scene Statistics (NSS). I first develop relevant theoretical bounds on the processing of foveated--more generally LSV-filtered (Linear Scale Variant)--signals, which provide a rigorous basis to linear post-processing operations performed on foveated images. The major contribution of this dissertation, however, lies in the discovery and elucidation of two major statistical characterizations of natural scenes and their subsequent deployment for devising optimal fixation strategies. The first is a novel characterization of the contrast statistics of natural scenes, parameterized by the eccentricity at which each contrast level is measured across the LSV-filtered image. This formulation of contrast statistics finds natural application in devising fixation patterns that optimally extract contrast information from the image. I further demonstrate that the resulting fixation patterns are nearly optimal in the sense of minimizing the global MSE of the LSV-filtered image. The second is the characterization of the non-stationary structure of natural images via the development of the concept of non-stationarity indices that measure the extent of non-stationarity across the image. The theoretical motivation of our approach lies in a novel characterization of image patch statistics I developed, called Multilinear Independent Component Analysis (MICA), wherein the statistical interactions between the pseudo-independent components are captured via a multilinear expansion of the joint probability density being modeled. This modeling technique enables the derivation of a theoretical measure of non-stationary in natural scenes that subsequently motivates computationally efficient non-stationarity indices--a variant of which is then deployed to furnish optimal texture-based fixations natural images. The fixation patterns generated by our information-theoretic approaches are quantitatively shown to match very well with human fixation patterns and offer considerable explanatory and predictive power over previously well-known fixation strategies. These results point the way towards a unified information-theoretic understanding of low-level fixation processes; and further demonstrate the importance of incorporating low-level visual information into visual search strategies--thereby providing a foundation upon which high-level visual information relating to scene context and object structures can be incorporated.Item Pattern recognition: using neural networks to classify digital modulated signals(Texas Tech University, 1989-05) Lodi, Shakeel IqbalArtificial neural systems are drawing increasing interest as useful tools for adaptive pattern recognition. The classification and subsequent recognition of digital modulated signals is a classical problem in pattern recognition. The ability to classify an incoming digital modulated signal accurately allows one to use an optimum detector to recover the transmitted message. The research focuses on using back propagation and Hogg & Hubermann learning algorithms for artificial neural networks in the demodulation process to classify three types of digital modulated signals: Amplitude Shift Keyed, Frequency Shift Keyed, and Phase Shift Keyed in the presence of zero-mean Gaussian noise. The principle difficulty encountered in pattern recognition problems is to present the patterns in a formalized manner. In many successful pattern recognition systems, a pattern is first normalized (e.g., aligned in position), then processed (e.g.. by feature extraction) and then classified. A signal preprocessor has been developed to perform all the front-end pattern processing, which allows the neural network to perform its tasks, such as association and processing of inexact knowledge. The output of the front-end signal preprocessor, a feature vector, is the input to the neural network model. The features currently being used are zero-crossings, Fourier Transforms, Power Spectral Density analysis, and Signal Slope analysis. A description of the signal preprocessor design and a discussion of the signal features used to train the neural networks is provided. Finally, the performance of the two neural network models is provided based on the following: recognition rate of untrained patterns and learning speed.Item Recognition of patterns in electronic communication signals using neural networks(Texas Tech University, 1991-05) Stubbendieck, Gregg T.This paper presents the results of research into automatic recognition of a class of electronic communication signals using a Back Propagation (BP) model neural network. Communication signals present an important and interesting pattern recognition challenge since they change unpredictably over time in accordance with the information they carry. There are situations in which a receiver has no prior knowledge of a particular signal and must classify it before interpreting it. The communication systems of interest here use frequency division techniques to multiplex several telegraph sub-signals in a standard communication channel. Previous research in recognizing these signals has demonstrated good recognition rates at the cost of expensive signal preprocessing. In this research, a BP network, smaller than networks used previously on this problem, was trained to recognize several types of these signals with a high degree of accuracy using a feature vector that is computationally less expensive and smaller than previous feature vectors. The observation that the BP network is tolerant of noise in patterns is reaffirmed in this research.Item Superresolution of real image sequence(Texas Tech University, 2004-12) Feng, ZhanpengImage superresolution has attracted substantial attention in the image processing community in recent years. Valuable techniques have been developed, and practical results have been obtained. However, in much of the literature, successes are frequently demonstrated in synthetic simulations, which limit a technique's practical use. This thesis will develop a technique to superresolve a real image sequence. This technique consists of three portions: system blur and noise removal, image registration, and sequence combination. First, the system blur and noise removal is achieved by a new approach of Point Spread Function (PSF) estimation. This approach is easy, cost-effective, and accurate compared to traditional methods. Then, image registration is performed, based on inserted fiducials. Translational shifts, rotation, scaling, and geometric distortions can be handled by this method. Finally, three different framecombining algorithms are implemented and compared. These techniques are demonstrated on an image sequence taken by a Canon EOS D30 digital camera. Quarter pixel superresolved images with sharper edges are obtained. The results confirm the effectiveness of these techniques. Analyses are done in terms of performance and implementation complexity.