It is often that images generated from Synthetic Aperture Radar (SAR) are noisy, distorted, or incomplete pictures of a target or target region. As the goal for most SAR research pertains to automatic target recognition (ATR), extensive filtering and image processing is required in order to extract the features necessary to carry out ATR. This thesis investigates the use of Artificial Neural Networks (ANNs) in order to improve upon the feature extraction process by laying the foundation for ANN SAR ATR algorithms and programs. The first technique investigated is that of an ANN edge detector designed to be invariant to multiplicative speckle noise. The algorithm designed uses the Back Propagation (BP) algorithm to teach a multi-layer perceptron network to detect edges. In order to do so, several parameters within a Sliding Window (SW), are calculated as the inputs to the ANN. The ANN then outputs an edge map that includes the outer edge features of the target as well as some internal edge features. The next technique that is examined is a pattern recognition and target reconstruction algorithm based off of the associative memory ANN known as the Hopfield Network (HN). For this version of the HN, the network is trained with a collection of varying geometric shapes. The output of the network is a nearest-fit representation of the incomplete image data input. Because of the versatility of this program, it is also able to reconstruct incomplete 3D models determined from SAR data. The final technique investigated is an automatic rotation procedure to detect the change in perspective relative to the platform. This type of detection can prove useful if used for target tracking or 3D modeling where the direction vector or relative angle of the target is a desired piece of information.

Computer vision is becoming a ubiquitous technology in many industries on account of its speed, accuracy, and long-term cost efficacy. The ability of a computer vision system to quickly and efficiently make quality decisions has made computer vision a popular technology on inspection lines. However, few companies in the agriculture industry use computer vision because of the non-uniformity of sellable produce. The small number of agriculture companies that do utilize computer vision use it to extract features for size sorting or for a binary grading system: if the piece of fruit has a certain color, certain shape, and certain size, then it passes and is sold. If any of the above criteria are not met, then the fruit is discarded. This is a highly wasteful and relatively subjective process. This thesis proposes a process to undergo to use computer vision techniques to extract features of fruit and build a model to predict shelf-life based on the extracted features. Fundamentally, the existing agricultural processes that do use computer vision base their distribution decisions on current produce characteristics. The process proposed in this thesis uses current characteristics to predict future characteristics, which leads to more informed distribution decisions. By modeling future characteristics, the process proposed will allow fruit characterized as “unfit to sell” by existing standards to still be utilized (i.e. if the fruit is too ripe to ship across the country, it can still be sold locally) which decreases food waste and increases profit. The process described also removes the subjectivity present in current fruit grading systems. Further, better informed distribution decisions will save money in storage costs and excess inventory. The proposed process consists of discrete steps to follow. The first step is to choose a fruit of interest to model. Then, the first of two experiments is performed. Sugar content of a large sample of fruit are destructively measured (using a refractometer) to correlate sugar content to a color range. This step is necessary to determine the end-point of data collection because stages of ripeness are fundamentally subjective. The literature is consulted to determine “ripe” sugar content of the fruit and the first experiment is undertaken to correlate a color range that corresponds to the “ripe” sugar content. This feature range serves as the end-point of the second experiment. The second experiment is large-scale data collection of the fruit of interest, with features being recorded every day, until the fruit reaches end-of-life as determined by the first experiment. Then, computer vision is used to perform feature extraction and features are recorded over each sample fruit’s lifetime. The recorded data is then analyzed with regression and other techniques to build a model of the fruit’s shelf-life. The model is finally validated. This thesis uses bananas as a proof of concept of the proposed process.

Yes / A confocal microscope provides a sequence of images of the various corneal layers and structures at different depths from which medical clinicians can extract clinical information on the state of health of the patient’s cornea. Preprocessing the confocal corneal images to make them suitable for analysis is very challenging due the nature of these images and the amount of the noise present in them. This paper presents an efficient preprocessing approach for confocal corneal images consisting of three main steps including enhancement, binarisation and refinement. Improved visualisation, cell counts and measurements of cell properties have been achieved through this system and an interactive graphical user interface has been developed.

No / Word spotting in continuous speech is considered a challenging issue due to dynamic nature of speech. Literature contains a variety of novel techniques for the isolated word recognition and spotting. Most of these techniques are based on pattern recognition and similarity measures. This paper amalgamates the use of different techniques that includes wavelet transform, feature extraction and Euclidean distance. Based on the acoustic features, the proposed system is capable of identifying and localizing a target (test) word in a continuous speech of any length. Wavelet transform is used for the time-frequency representation and filtration of speech signal. Only high intensity frequency components are passed to feature extraction and matching process resulting robust performance in terms of matching as well as computational cost.

This thesis is concerned with model-based non-rigid registration of single-modality magnetic resonance images of compressed and uncompressed breast tissue in breast cancer diagnostic/interventional imaging. First, a volumetric registration algorithm is developed which solves the registration as a state estimation problem. Using a static deformation model. To reduce computations, the similarity measure is calculated at some specific points called control points. These control points can be from a low resolution image grid or any irregular image grid. Our numerical analysis has shown that control points placed in the area without much information; i.e with small or no changes in image intensity, yield negligible deformation. Therefore, the selection of the control points can significantly impact the accuracy and computation complexity of the registration algorithms. An extension of the speeded up robust features (SURF) to 3D is proposed for enhanced selection of the control points in deformable image registration. The impact of this new control point selection method on the performance of the registration algorithm is analyzed by comparing it to the case where regular grid control points are used. The results show that the number of control points could be reduced by a factor of ten with new selection methodology without sacrificing performance. Second image registration method is proposed in which, based on a segmented pre-operative image, a deformation model of the breast tissue is developed and discretized in the spatial domain using the method of finite elements. The compression of the preoperative image is modeled by applying smooth forces on the surface of the breast where compression plates are placed. Image registration is accomplished by formulating and solving an optimization problem. The cost function is a similarity measure between the deformed preoperative image and intra-operative image computed at some control point and the decision variables are the tissue interaction forces. / Thesis / Master of Applied Science (MASc)

It is a popular practice in the remote-sensing community to apply principal component analysis (PCA) on a higher-dimensional feature space to achieve dimensionality-reduction. Several factors that have led to the popularity of PCA include its simplicity, ease of use, availability as part of popular remote-sensing packages, and optimal nature in terms of mean square error. These advantages have prompted the remote-sensing research community to overlook many limitations of PCA when used as a dimensionality-reduction tool for classification and target-detection applications. This thesis addresses the limitations of PCA when used as a dimensionality-reduction technique for extracting discriminating features from hyperspectral data. Theoretical and experimental analyses are presented to demonstrate that PCA is not necessarily an appropriate feature-extraction method for high-dimensional data when the objective is classification or target-recognition. The influence of certain data-distribution characteristics, such as within-class covariance, between-class covariance, and correlation on PCA transformation, is analyzed in this thesis. The classification accuracies obtained using PCA features are compared to accuracies obtained using other feature-extraction methods like variants of Karhunen-Loève transform and greedy search algorithms on spectral and wavelet domains. Experimental analyses are conducted for both two-class and multi-class cases. The classification accuracies obtained from higher-order PCA components are compared to the classification accuracies of features extracted from different regions of the spectrum. The comparative study done on the classification accuracies that are obtained using above feature-extraction methods, ascertain that PCA may not be an appropriate tool for dimensionality-reduction of certain hyperspectral data-distributions, when the objective is classification or target-recognition.

No description available.

No description available.

No description available.

No description available.