Dissertations / Theses on the topic 'Canny's edge detection method'

This diploma thesis deals with the issue of the recognition of road signs in the video sequence. Such systems increase the traffic safety and are implemented by major car factories in the manufactured cars (Opel, BMW). First, the motivation for the utilisation of these systems is presented, followed by the survey of the current state of the art methods. Finally, a specific road-sign detection method is chosen and described in detail. The method uses advanced techniques of signal processing. Segmentation method in color space is used for sign detection and subsequent classification is accomplished by linear classification with optional use of PCA method. In addition, the method contains the prediction of road sign positions based on Kalman filtering. Implemented system yields relatively accurate results and overall analysis and discussion is enclosed.

<p>This thesis presents a 3D semi-automatic segmentation technique for extracting the lumen surface of the Carotid arteries including the bifurcation from 3D and 4D ultrasound examinations. </p><p>Ultrasound images are inherently noisy. Therefore, to aid the inspection of the acquired data an adaptive edge preserving filtering technique is used to reduce the general high noise level. The segmentation process starts with edge detection with a recursive and separable 3D Monga-Deriche-Canny operator. To reduce the computation time needed for the segmentation process, a seeded region growing technique is used to make an initial model of the artery. The final segmentation is based on the inflatable balloon model, which deforms the initial model to fit the ultrasound data. The balloon model is implemented with the finite element method. </p><p>The segmentation technique produces 3D models that are intended as pre-planning tools for surgeons. The results from a healthy person are satisfactory and the results from a patient with stenosis seem rather promising. A novel 4D model of wall motion of the Carotid vessels has also been obtained. From this model, 3D compliance measures can easily be obtained.</p>

Event Detection is a process of identifying terrain flatness from which localized events such as potholes in the terrain surface can be found and is an important tool in pavement health monitoring and vehicle performance inspection. Repeated detection of terrain surfaces over an extended period of time can be used by highway engineers for long term road health monitoring. An accurate terrain map can allow maintenance personnel for identifying deterioration in road surface for immediate correction. Additionally, knowledge of the events in terrain surface can be used to predict the performance the vehicles would experience while traveling over it. Event detection is composed of two processes: event edging and stitching edges to events. Edge detection is a process of identifying significant localized changes in the terrain surface. Many edge detection methods have been designed capable of capturing edges in terrain surfaces. Gradient searches are frequently used in image processing to recover useful information from images. The issue with using a gradient search method is that it returns deterministic values resulting in edges which are less precise. In order to predict the precision of the terrain surface, the individual nodal probability densities must be quantified and finally combined for the precision of terrain surface. A Comparative Nodal Uncertainty Method is developed in this work to detect edges based on the probability distribution of the nodal heights within some local neighborhood. Edge stitching is developed to group edges to events in a correct sequence from which an event can be determined finally.<br>Master of Science

Epileptic seizure presents a formidable threat to the life of its sufferers, leaving them unconscious within seconds of its onset. Having a mortality rate that is at least twice that of the general population, it is a true cause for concern which has gained ample attention from various research communities. About 800 million people in the world will have at least one seizure experience in their lifespan. Injuries sustained during a seizure crisis are one of the leading causes of death in epilepsy. These can be prevented by an early detection of seizure accompanied by a timely intervention mechanism. The research presented in this dissertation explores Kriging methods to exploit spatial correlations of electroencephalogram (EEG) Signals from the brain, for fast and accurate seizure detection in the Internet of Medical Things (IoMT) using edge computing paradigms, by modeling the brain as a three-dimensional spatial object, similar to a geographical panorama. This dissertation proposes basic, hierarchical and distributed Kriging models, with a deep neural network (DNN) wrapper in some instances. Experimental results from the models are highly promising for real-time seizure detection, with excellent performance in seizure detection latency and training time, as well as accuracy, sensitivity and specificity which compare well with other notable seizure detection research projects.

В [1] описан метод локального распознавания контуров, который основан на использовании функции непропорциональности по производной первого порядка для функций, заданных параметрически [2,3]. Растровый формат, в котором часто представлены распознаваемые контуры, затрудняет построение параметрической зависимости в полярной системе координат из-за вносимых им погрешностей. В [1] для решения этой задачи применяется линейная аппроксимация фрагментов контурных изображений. Такой подход позволяет уменьшить погрешность, вносимую растровым форматом, а также производить более эффективный контроль пересечений контуров радиус-векторами.

Les ``C-arm'' sont des systémes de radiologie interventionnelle fréquemment utilisés en salle d'opération ou au lit du patient. Des images 3D des structures anatomiques internes peuvent être calculées à partir de multiples radiographies acquises sur un ``C-arm mobile'' et isocentrique décrivant une trajectoire généralement circulaire autour du patient. Pour cela, la géométrie conique d'acquisition de chaque radiographie doit être précisément connue. Malheureusement, les C-arm se déforment en général au cours de la trajectoire. De plus leur motorisation engendre des oscillations non reproductibles. Ils doivent donc être calibrés au cours de l'acquisition. Ma thèse concerne la calibration intrinsèque d'un C-arm à partir de la détection de la projection du collimateur de la source dans les radiographies.Nous avons développé une méthode de détection de la projection des bords linéaires du collimateur. Elle surpasse les méthodes classiques comme le filtre de Canny sur données simulées ou réelles. La précision que nous obtenons sur l'angle et la position (phi,s) des droites est de l'ordre de: phi{RMS}=+/- 0.0045 degrees et s{RMS}=+/- 1.67 pixels. Nous avons évalué nos méthodes et les avons comparés à des méthodes classiques de calibration dans le cadre de la reconstruction 3D<br>Mobile isocentric x-ray C-arm systems are an imaging tool used during a variety of interventional and image guided procedures. Three-dimensional images can be produced from multiple projection images of a patient or object as the C-arm rotates around the isocenter provided the C-arm geometry is known. Due to gravity affects and mechanical instabilities the C-arm source and detector geometry undergo significant non-ideal and possibly non reproducible deformation which requires a process of geometric calibration. This research investigates the use of the projection of the slightly closed x-ray tube collimator edges in the image field of view to provide the online intrinsic calibration of C-arm systems.A method of thick straight edge detection has been developed which outperforms the commonly used Canny filter edge detection technique in both simulation and real data investigations. This edge detection technique has exhibited excellent precision in detection of the edge angles and positions, (phi,s), in the presence of simulated C-arm deformation and image noise: phi{RMS} = +/- 0.0045 degrees and s{RMS} = +/- 1.67 pixels. Following this, the C-arm intrinsic calibration, by means of accurate edge detection, has been evaluated in the framework of 3D image reconstruction

This thesis proposes a method to reconstruct a frontal facial video basedon encoding done with the facial profile of another video sequence.The reconstructed facial video will have the similar facial expressionchanges as the changes in the profile video. First, the profiles for boththe reference video and for the test video are captured by edge detection.Then, asymmetrical principal component analysis is used to model thecorrespondence between the profile and the frontal face. This allows en-coding from a profile and decoding of the frontal face of another video.Another solution is to use dynamic time warping to match the profilesand select the best matching corresponding frontal face frame for re-construction. With this method, we can reconstructed the test frontalvideo to make it have the similar changing in facial expressions as thereference video. To improve the quality of the result video, Local Lin-ear Embedding is used to give the result video a smoother transitionbetween frames.

This thesis is focused on face detection in static picture. Theoretical part contains color spaces (RGB, HSI, YCbCr), methods for skin detection (explicit, parametric or non-parametric methods), image metric, edge detection, mathematical morphology, methods for classification faces (appearance-based methods, feature invariant approaches, knowledge-based methods, template matching methods). Practical part of this thesis contains concept and practical realization two algorithms for segmentation skin in static image (simple method based on Cr chroma components and statistical method). Practical part contains concept and practical realization two algorithms for classification face (appearance-based method and template matching method) too.

This work deal with determination topohraphic parameters of a randomly rough surface by the help of method of shearing interferometry. It is a optical method for determination surface roughness. The basic idea is based of on deformation interference strips which are made by interference of the same mutually translated monochrome luminous wavefronts. The wavefront is created after transit or reflection monochrome lights from the surface of a studied sample. The wavefronts creates picture with deformed interference strips , which carries information about character of the surface. This information can be profited by algorithms of image processing from the picture . The thesis was developed in research project MSM 0021630529 Intelligent Systems in Automation.

Bei strukturorientierten, aus relativ großen Holzpartikeln aufgebauten Holzwerkstoffen wie z.B. OSB (oriented strand board) addieren sich die gerichteten Festigkeiten der einzelnen Lagen je nach Orientierung der Partikel und der Verteilung ihrer Größenparameter. Wünschenswert wäre eine Messung der Partikelgeometrie und Orientierung möglichst im Prozess, z.B. am Formstrang vor der Presse direkt durch den „Blick auf das Vlies“. Bisher sind regelmäßige on-line-Messungen der Spangeometrie aber nicht möglich, und Einzelspanmessungen werden nicht vorgenommen, weil sie zu aufwändig wären. Um die Partikelkonturen zunächst hinreichend für die Vermessung zu restaurieren und dann zu vermessen, muss ein mehrstufiges Verfahren angewendet werden, das eine Szene mit Strands und mehr oder weniger deutlichen Kanten zunächst als „Grauwertgebirge“ auffasst. Zur Segmentierung reicht ein Watershed-Algorithmus nicht aus. Auch ein zweistufiger Kantendetektor nach Canny liefert allein noch kein ausreichendes Ergebnis, weil sich keine geschlossenen Objektkonturen ergeben. Hinreichend dagegen ist ein komplexes Verfahren auf der Grundlage der Höhenschichtzerlegung und nachfolgenden Synthese: Nach einer Transformation der Grauwerte des Bildes in eine reduzierte, gleichverteilte Anzahl von Höhenschichten werden zwischen diesen die lokalen morphologischen Gradienten berechnet und herangezogen für die Rekonstruktion der ursprünglichen Spankonturen. Diese werden aus den Höhenschichten aufaddiert, wobei allerdings nur Teilflächen innerhalb eines für die gesuchten Spangrößen plausiblen Größenintervalls einbezogen werden, um Störungen zu unterdrücken. Das Ergebnis der Rekonstruktion wird zusätzlich verknüpft mit den bereits durch einen Canny-Operator im Originalbild detektierten deutlichen Kanten und morphologisch bereinigt. Diese erweiterte Höhenschichtanalyse ergibt ausreichend segmentierte Bilder, in denen die Objektgrenzen weitgehend den Spankonturen entsprechen. Bei der nachfolgenden Vermessung der Objekte werden Standard-Algorithmen eingesetzt, wobei sich die Approximation von Spankonturen durch momentengleiche Ellipsen als sinnvoll erwies. Verbliebene Fehldetektionen können bei der Vermessung unterdrückt werden durch Formfaktoren und zusätzliche Größenintervalle. Zur Darstellung und Charakterisierung der Größenverteilungen für die Länge und die Breite wurden die nach der Objektfläche gewichtete, linear skalierte Verteilungsdichte (q2-Verteilung), die Verteilungssumme und verschiedene Quantile verwendet. Zur Umsetzung und Demonstration des Zusammenwirkens der verschiedenen Algorithmen wurde auf der Basis von MATLAB das Demonstrationsprogramm „SizeBulk“ entwickelt, das Bildfolgen verarbeiten kann und mit dem die verschiedenen Varianten der Bildaufbereitung und Parametrierung durchgespielt werden können. Das Ergebnis des Detektionsverfahrens enthält allerdings nur die vollständigen Konturen der ganz oben liegenden Objekte; Objekte unterhalb der Außenlage sind teilweise verdeckt und können daher nur unvollständig vermessen werden. Zum Test wurden daher synthetische Bilder mit vereinzelten und überlagerten Objekten bekannter Größenverteilung erzeugt und dem Detektions- und Messverfahren unterworfen. Dabei zeigte sich, dass die Größenstatistiken durch den Überlagerungseffekt und auch die Spanorientierung zwar beeinflusst werden, dass aber zumindest die Modalwerte der wichtigsten Größenparameter Länge und Breite meist erkennbar bleiben. Als Versuchsmaterial dienten außer den synthetischen Bildern verschiedene Sortimente von OSB-Strands aus Industrie- und Laborproduktion. Sie wurden sowohl manuell vereinzelt als auch zu einem Vlies arrangiert vermessen. Auch bei realen Strands zeigten sich gleiche Einflüsse der Überlagerung auf die Größenverteilungen wie in der Simulation. Es gilt aber auch hier, dass die Charakteristika verschiedener Spankontingente bei gleichen Aufnahmebedingungen und Auswerteparametern gut messbar sind bzw. dass Änderungen in der gemessenen Größenverteilung eindeutig den geometrischen Eigenschaften der Späne zugeordnet werden können. Die Eignung der Verarbeitungsfolge zur Charakterisierung von Spangrößenverteilungen bestätigte sich auch an Bildern, die ausschließlich am Vlies auf einem Formstrang aufgenommen wurden. Zusätzlich wurde nachgewiesen, dass mit der erweiterten Höhenschichtanalyse auch Bilder von Spanplattenoberflächen ausgewertet werden könnten und daraus auf die Größenverteilung der eingesetzten Deckschichtspäne geschlossen werden kann. Das vorgestellte Verfahren ist daher eine gute und neuartige Möglichkeit, prozessnah an Teilflächen von OSB-Vliesen anhand von Grauwertbildern die Größenverteilungen der Strands zu charakterisieren und eignet sich grundsätzlich für den industriellen Einsatz. Geeignete Verfahren waren zumindest für Holzpartikel bisher nicht bekannt. Diese Möglichkeit, Trends in der Spangrößenverteilung automatisch zu erkennen, eröffnet daher neue Perspektiven für die Prozessüberwachung<br>The strength of wood-based materials made of several layers of big and oriented particles like OSB (oriented strand board) is a superposition of the strengths of the layers according to the orientation of the particles and depending from their size distribution. It would be desirable to measure particle geometry and orientation close to the production process, e.g. with a “view onto the mat”. Currently, continuous on-line measurements of the particle geometry are not possible, while measurements of separated particles would be too costly and time-consuming. Before measuring particle shapes they have to be reconstructed in a multi-stage procedure which considers an image scene with strands as “gray value mountains”. Segmentation using a watershed algorithm is not sufficient. Also a two-step edge detector according to Canny does not yield closed object shapes. A multi-step procedure based on threshold decomposition and recombination however is successful: The gray values in the image are transformed into a reduced and uniformly distributed set of threshold levels. The local morphological gradients between these levels are used to re-build the original particle shapes by adding the threshold levels. Only shapes with a plausible size corresponding to real particle shapes are included in order to suppress noise. The result of the reconstruction from threshold levels is then matched with the result of the strong edges in the original image, which had been detected using a Canny operator, and is finally cleaned with morphological operators. This extended threshold analysis produces sufficiently segmented images with object shapes corresponding extensively to the particle shapes. Standard algorithms are used to measure geometric features of the objects. An approximation of particle shapes with ellipses of equal moments of inertia is useful. Remaining incorrectly detected objects are removed by form factors and size intervals. Size distributions for the parameters length and width are presented and characterized as density distribution histograms, weighted by the object area and linearly scaled (q2 distribution), as well as the cumulated distribution and different quantiles. A demonstration software “SizeBulk” based on MATLAB has been developed to demonstrate the computation and the interaction of algorithms. Image sequences can be processed and different variations of image preprocessing and parametrization can be tested. However, the detection procedure yields complete shapes only for those particles in the top layer. Objects in lower layers are partially hidden and cannot be measured completely. Artificial images with separated and with overlaid objects with a known size distribution were generated to study this effect. It was shown that size distributions are influenced by this covering effect and also by the strand orientation, but that at least the modes of the most important size parameters length and width remain in evidence. Artificial images and several samples with OSB strands from industrial and laboratory production were used for testing. They were measured as single strands as well as arrangements similar to an OSB mat. For real strands, the same covering effects to the size distributions revealed as in the simulation. Under stable image acquisition conditions and using similar processing parameters the characteristics of these samples can well be measured, and changes in the size distributions are definitely due to the geometric properties of the strands. The suitability of the processing procedure for the characterization of strand size distributions could also be confirmed for images acquired from OSB mats in a production line. Moreover, it could be shown that the extended threshold analysis is also suitable to evaluate images of particle board surfaces and to draw conclusions about the size distribution of the top layer particles. Therefore, the method presented here is a novel possibility to measure size distributions of OSB strands through the evaluation of partial gray value images of the mat surface. In principle, this method is suitable to be transferred to an industrial application. So far, methods that address the problem of detecting trends of the strand size distribution were not known, and this work shows new perspectives for process monitoring

Nous étudions dans cette thèse la possibilité de reconnaitre des objets dans des images compressées, sans les reconstruire. L'algorithme de compression le plus adapte semble celui fonde sur l'extraction des contours multi échelle quinconce des images. Le problème de la reconnaissance nous amène à introduire un nouvel outil de comparaison d'images binaires: la distance de Hausdorff censurée. Cet outil s'est avéré robuste et rapide à calculer. Ces deux points sont étudiés avec soin. Cette distance est enfin utilisée pour reconnaitre et localiser des objets spécifiques dans des scènes de grande taille. Nous proposons trois approches multi échelles pour résoudre ce problème, qui prennent en compte le fait que l'objet recherché peut être en partie caché, ou qu'il peut être vu sous un angle différent de son modèle. L'algorithme que nous avons développé est rapide sur station de travail classique. Sa robustesse a été étudiée soigneusement. Sa parallélisation nous permet d'atteindre le temps réel dans un cadre opérationnel raisonnable

Reconstruction tomographique de la distribution radioactive d'un traceur placé dans le milieu non atténuant puis atténuant. Différentes méthodes de reconstruction, performances et contrôle de la qualité du détecteur, dégradation des images. Méthode de détection automatique des contours

碩士<br>逢甲大學<br>資訊工程學系<br>103<br>Edge detection in 3D depth images is more vital and useful than that in 2D images for the following intermediate-level and high-level vision, such as segmentation and shape reconstruction. Kinect, a lower-cost depth camera, it has been concern for many scholars. But depth image is often not very accurate and stable due to depth holes and inconsistent depth values. This paper in order to improve the effectiveness of Kinect image processing, will study the edge detection for depth image. Now, the technology of edge detection has been very ripe, and Canny edge detection is the most popular. Canny edge detection computed the gradient magnitude and direction of the pixels, and used hysteresis thresholding to continuous edge curve. This paper will notice the result of using Canny edge detecton on depth image. We can found the result wasn’t good. Because of the gradient magnitude at the depth image was too small. Therefore, this paper will study this problem, propose a new approache. First, we rewrite the step of compute gradient which is in Canny edge detection. Then, we use Canny edge detection to detect edge from RGB image and depth image. Third, found the Roof edge of the depth image as the edge candidate. At last, we use the RGB image’s edge to locate the real edge. If the pixel is RGB image’s edge and edge candidate, this pixel is real edge.Then we make up the depth edge and get a new edge image. Our method can useful to detect depth edges accurately and provide effective assistance for other research of depth image.

碩士<br>國立臺灣師範大學<br>機電科技研究所<br>99<br>Glottis servers a very major function in animals and its dynamic change is also a critical subject in physiological research. This research aims to analyze the image processing methods for glottis by using Matlab computer programming tools to estimate the glottal area and to plot the correspondent charts. We hope to simplify the manual processing of targeting glottal area in digital image and the calculation of area. The study employs Canny edge detection method and morphological segmentation method to make the curves of glottal area. The results show that the algorithm adopted by our experiments could successfully locate the edge of glottal area and form a closed-end contour to obtain the measure of glottal area. Not only would the method help further analyses, but they create higher efficiency than does the traditional active contour liner model.

碩士<br>國立虎尾科技大學<br>資訊管理研究所<br>95<br>Edge is one of common features in images. By taking the complete information of edges in an image is to generate the sharp profile of the object. Watermarking is a technique that plays a role as the protector for the image’s copyright information. In this thesis, we propose a robust watermarking technique based on edge features in vector quantization (VQ) index table and the concept of association rules in data mining. Besides, we also present two edge detection approaches using a novel sliding window technique. The first scheme is to define 4-itemset association rules in VQ index table to explore some edge characteristics such that association rules of a watermark and an original image can be obtained, respectively. Subsequently, by embedding association rules of a watermark into association rules of an original image, the purpose for protecting intellectual property rights is achieved. The second scheme is to apply sliding window technique to explore the variations of image pixels for edge detection. All pixels in an image will be included into different sub-windows with different size according to the heterogeneities of edges. After performing this concept on image pixels to explore edge features from four variant directions, four different directional edge maps are generated. Afterwards, eight edge selectors are utilized to sieve out possible edges, and the final edge image is determined by voting. The third scheme also applies the same concept, sliding window technique with dynamic size, to propose an edge detection approach and improve the second scheme. Along different sliding direction, the window size will be modified according to the variation of a pixel and its neighboring pixels in the same sliding direction. This concept is utilized to generate eight directional edge maps from eight different sliding directions. Subsequently, directional edge candidates can be obtained by regarding some pixels in those edge maps as edges, and from each directional edge candidate, some connective edge candidates will be considered to be edges so as to generate a directional edge rough. The final edge image is created by performing the basic logical operation, OR, to some directional edge roughs. Relevant experimental results demonstrate that our proposed robust watermarking technique can survive geometric operation attacks, and any a complex watermark with size larger than an original image also can be implemented. Besides, the other two schemes applied to detect edges also can achieve better results with less double edges and speckles occurring.

碩士<br>國立成功大學<br>電腦與通信工程研究所<br>102<br>In traditional stereo matching method, global method is more accurate but spending more time, and have more correct rate in occlusion area. On the contrary, local method is usually fast but have bad performance, and easily is influence by noise. This paper proposed a novel method to compute disparity between two images. It is based on local method, but its cost be aggregated in like-global way. This aggregation is processed by a weight map which created by bilateral filter concept. Every pixel transfer its own cost information to all pixels on the same object, but this information would be restricted by weight map. After finishing preliminary depth map, we use L-R check to find occlusion and mismatch pixel. Then fix occlusion by the smallest disparity nearby. At last, we use bilateral filter clean up whole depth map. All of above computing process can be parallelized on GPU machine or cloud sever. Although this algorithm is designed for low-level machine, it still exerts high performance in the world of high-level hardware.

碩士<br>逢甲大學<br>自動控制工程學系<br>102<br>This research study about the combination of dynamic image edge detection and specified pronunciation keyboard. It can provide people with disabilities or people with Amyotrophic Lateral Sclerosis through specified pronunciation keyboard to writing article, surfing Internet, and so on. The patients can express their autonomy needs,and taking into account the state of the patients are difficult to manipulate the keyboard accurately which the general market to sell. The design of this research is that users only use phonetic symbols to represent the phrase. And typing action don&;#39;t need to select the word with a full section of the complete text to communicate. So it can improve the speed of communication. Then the system will be to identify screenshot of hand image. Use K-means clustering algorithm to find the baseline of the hand. To identify the change volume of degree in the angle of the hand baseline by the fuzzy algorithm. And establishing the membership function of fuzzy to observe the degree of finger swing. According to the situation of people with disability , the activity of hands are differences. When the swing angle of hands is too small,which is below mouse button clicking action of the Initial standard angle. When the swing angle of hands cumulative three times to lower than the standard angle, the system will reduce the standard of the initial angle. Let the click angle of mouse button clicking action to fit users. After the angle adjusting , the clicking action will cumulative three times lower than the initial angle. It means that new adjust angle does not apply to the user, the system will adjust the angle again. Then adjusting the best angle which is apposite angle of the user&;#39;s hand swing. Using this system on a computer screen instead of that the use of electronic communication board requires two or more people convenientiy. So this reaserch propose the secified pronunciation interface.

碩士<br>淡江大學<br>電機工程學系碩士班<br>93<br>Many video compression standards, such as MPEG-4 and MPEG-7, have supported object-based multimedia coding that allows user to interact, search and exchange the objects in the images or video sequences. For supporting these features, the object segmentation and tracking in the video sequences play an essential and important role. This thesis proposes a solution algorithm to track one or multiple moving objects in frames of a video sequence, including edge detection algorithm, image segmentation algorithm and trajectory estimation functions. The object segmentation algorithm proposed in this thesis is based on exploring contour of the image. Therefore, to extract the desired objects with more precisely, an effective method for extracting the contour of the image is needed. The conventional edge detection algorithm is no longer satisfy there requirements. For example, Sobel’s edge detector can successfully sketch out the apparently contour of the objects. However, most of the thin edges in the image normally be eliminated by Sobel’s edge detector. The DoE (difference of exponential) method is able to track the stronger edges of the image, but the performance of extracting thin edges is not acceptable. However, in many cases, thin edges represent important features in the image, and should not be eliminated or discarded. This thesis presents a novel mathematical morphology based edge detector to enhance the performance of extracting thin edges in a still image. According to the mean value and standard derivation of the pixel in the image, the proposed method can enhance thin edges in the image for extracting by a global threshold value. After the edge detection process, this thesis proposes for applying a novel object segmentation algorithm to the image for extracting objects of the image. There are several popular algorithm had been developed for image segmentation, such as snake energy model and watershed algorithm. For snake energy model, it requires a manually-drawn initial snake and adjusts weighting parameters in the snake model. The snake is controlled by two energies, which are internal energy and external energy. The snake iterations are converged when these two energies reach to a balanced state. As to the watershed algorithm, it has the drawback of over-segmentation problem. This thesis presents a novel edge-based image segmentation algorithm that is capable of performing global image segmentation or segmenting desired objects in an image. The proposed algorithm provides more effective segmentation result than other methods by region growing method. A snake-energy-like cost function is developed to control the growing process for the algorithm to produce better segmentation results. While the growing phase is completed, the algorithm combines homogeneous regions together to extract more meaningful image objects. The segmentation algorithm proposed in this thesis is initialized by planting growing seeds into the image. Therefore, to extent our algorithms to video object tracking, this study proposes a scheme based on the previously segmentation result for automatically planting growing seeds into following video frames to extract the same objects on the following frames by the proposed scheme. Experimental results show that the proposed algorithms produce good performance on object segmentation and tracking.

碩士<br>雲林科技大學<br>電機工程系碩士班<br>98<br>Related researches and applications of License Plate Recognition (LPR) have been proceeding for decades. But, in practice, the license plate detection rate is usually error-prone to various backgrounds, illuminations, or skews. So is the subsequent character recognition rate. This is why most of the license plate recognition system just can work well under specific circumstances. Specifically, License plate detection is the fundamental step in LPR. In general, the license plate region is usually detected by the strong vertical edge feature of its interior characters. However, in the external environment, many backgrounds always have strong vertical edges as the license plate. This paper focuses on enlarging the vertical edge differentation between the license plate region and the non-plate region, that is, focuses on enhancing the vertical edge density and strength of the license plate region and weakening those of the non-plate region, e.g. weakening those of the grille. Therefore, the deblurred method of Wiener deconvolution is proposed to enhance and increase the vertical edges of the license plate. Then, 2-level 2D Discrete Wavelet Transform (DWT) is adopted to run the projection histogram of vertical edges. Furthermore, with the first-order local recursive Otsu segmentation, mathematical morphology, and edge density verification method, the license plate region can be detected smoothly. Experimental results show LPD system based on the proposed Wiener-deconvolution vertical edge enhancement method and 2-level 2D DWT, can achieve a much higher hit rate. After extracting the license plate region smoothly, the optical character recognition library of Tesseract OCR can be applied to recognize the license plate characters. On the other hand, this thesis implements a LPR with the Wiener-like vertical edge enhancement method onto Android embedded platform. The implementation result also verifies the Wiener-like vertical edge enhancement method is effective and feasible.

Ocular fundus image can provide information on pathological changes caused by local ocular diseases and early signs of certain systemic diseases. Automated analysis and interpretation of fundus images has become a necessary and important diagnostic procedure in ophthalmology. Among the features in ocular fundus image are the optic disc, fovea (central vision area), lesions, and retinal vessels. These features are useful in revealing the states of diseases in the form of measurable abnormalities such as length of diameter, change in color, and degree of tortuosity in the vessels. In addition, retinal vessels can also serve as landmarks for image-guided laser treatment of choroidal neovascularization. Thus, reliable methods for blood vessel detection that preserve various vessel measurements are needed. In this paper, we will examine the pathological issues in the analysis of retinal vessels in digital fundus images and give a survey of current image processing methods for extracting vessels in retinal images with a view to categorize them and highlight their differences and similarities. We have also implemented two major approaches using matched filter and mathematical morphology respectively and compared their performances. Some prospective research directions are identified.<br>Singapore-MIT Alliance (SMA)

University of Technology, Sydney. Faculty of Information Technology.<br>Automatic Number Plate Recognition (ANPR) is an important Intelligent Transportation System (ITS) technology, which distinguishes each vehicle as unique by recognising the characters in their number plates via image analysis and pattern recognition techniques. In an ANPR system, the most crucial part is number plate detection. The research presented in this thesis focuses on the detection mechanism and will rely on a third-party Optical Character Recognition (OCR) software for character recognition. Number Plate Detection (NPD) is a well-explored problem with many successful solutions. Although most of these solutions are reasonably fast and robust, they can be further improved to make them even faster and more robust to deal with various complex conditions in real-time. This thesis first presents a region-based NPD algorithm, which provides much more accurate detection results than previous NPD algorithms and is robust against interference characters in images. Then, a fast and robust edge-based NPD algorithm is developed. Tins algorithm can detect various number plates under various conditions in real-time with a high detection rate and a very low false positive rate. Similar work has not been reported elsewhere. Besides character information, the colour information of number plates also plays an important role in identifying each number plate as unique. Hence, this thesis also develops algorithms for classifying number plate colours. Histogram-based image matching methods are investigated, and a Gaussian Weighted Histogram Intersection (GWHI) algorithm is presented. This algorithm is shown to be much more robust against various colour variations than previous methods. Furthermore, a novel Colour Edge Co-occurrence Histogram (CECH) method is presented. It is shown to be particularly applicable for rapidly matching compound objects, such as number plates. Finally, histogram-based image matching technique on a hexagonal image structure is investigated. Gevers' idea of using Colour Ratio Gradient (CRG) for robust object matching is redefined on hexagonal structure, arid a novel Symmetric Colour Ratio Gradient (SCRG) method is developed. Experimental results demonstrate that the proposed SCRG method outperforms the Gevers’ CRG method. More contributions can be found in the appendices. A new virtual hexagonal structure is proposed, on which the time used for mapping a square-based image to hexagon-based image is dramatically reduced. Two basic image transformation operations and a novel edge detection algorithm are performed on the new structure. The results obtained in this thesis can also be applied to many other areas such as Character Detection, Text Detection, and Image/Video Retrieval