Abstract Pattern recognition and registration are integral elements of computer vision, which considers image patterns. This thesis presents novel blur, and combined blur and geometric invariant features for pattern recognition and registration related to images. These global or local features are based on the Fourier transform phase, and are invariant or insensitive to image blurring with a centrally symmetric point spread function which can result, for example, from linear motion or out of focus. The global features are based on the even powers of the phase-only discrete Fourier spectrum or bispectrum of an image and are invariant to centrally symmetric blur. These global features are used for object recognition and image registration. The features are extended for geometrical invariances up to similarity transformation: shift invariance is obtained using bispectrum, and rotation-scale invariance using log-polar mapping of bispectrum slices. Affine invariance can be achieved as well using rotated sets of the log-log mapped bispectrum slices. The novel invariants are shown to be more robust to additive noise than the earlier blur, and combined blur and geometric invariants based on image moments. The local features are computed using the short term Fourier transform in local windows around the points of interest. Only the lowest horizontal, vertical, and diagonal frequency coefficients are used, the phase of which is insensitive to centrally symmetric blur. The phases of these four frequency coefficients are quantized and used to form a descriptor code for the local region. When these local descriptors are used for texture classification, they are computed for every pixel, and added up to a histogram which describes the local pattern. There are no earlier textures features which have been claimed to be invariant to blur. The proposed descriptors were superior in the classification of blurred textures compared to a few non-blur invariant state of the art texture classification methods.

A biodiversidade das espécies existentes no riquíssimo reino vegetal, tornam os modelos tradicionais de taxonomia uma tarefa muito complexa e morosa, na qual o processo de classificação é tradicionalmente realizado manualmente. As dificuldades presentes nesse processo implicam na existência de poucas pesquisas de classificação vegetal utilizando métodos matemáticos e computacionais. Desta forma, visando contribuir com as técnicas de taxonomia já desenvolvidas, este estudo objetiva desenvolver e testar uma metodologia computacional de identificação de espécies vegetais por meio da análise da textura foliar. Motivado pelo projeto TreeVis, este trabalho realiza uma revisão dos métodos utilizados para análise de textura em imagens digitais (foco concentrado em extração de características e classificação), investigando a aplicabilidade de métodos tradicionais como matrizes de coocorrência, técnicas estado da arte como Gabor wavelets e também de novos e promissoras técnicas de análise de textura, como a dimensão fractal volumétrica. No contexto de classificação investiga-se métodos para reconhecimento de padrões lineares com base em análise de dados multivariados, não lineares com base na teoria das Redes Neurais Artificiais e métodos simples para combinação de diferentes classificadores (comitê de máquinas). Apesar da alta similaridade entre classes e similaridade intraclasses não adequada, os resultados alcançados mostraram-se excelentes. A melhor estratégia de classificação, utilizando comitê de máquinas com descritores de Gabor wavelets/cor e dimensão fractal volumétrica/cor, obteve uma probabilidade de acerto global de 96:32% nas 40 classes estudadas. Esse resultado demonstra como os métodos computacionais de análise de imagens, em especial análise de textura, podem contribuir facilitando e agilizando a tarefa de identificação de espécies vegetais / Biodiversity of species existing in the plant kingdom make the use of traditional models of taxonomy, a process of classification traditionally performed manually, a very complex and time-consuming task. Most of difficulties in that process result from the existence of few researches on plant classification using mathematical and computational methods. In this way, to contribute with the taxonomy techniques already developed, this study aims to develop and test a computational method for identifying plant species by leaf texture analysis. Motivated by the TreeVis project, this work is a comprehensive revision of texture analysis methods used in digital images (focus concentrated in features extraction and classification). This study investigates the applicability of traditional methods such as co-occurrence matrix, state of the art techniques as Gabor wavelets, and new and promising texture analysis methods, such as volumetric fractal dimension. In classification context is investigated methods of pattern recognition based on multivariate data analysis, artificial neural networks and committee machines. Although leaf classes present high similarity between classes and not appropriate similarity intraclasses, the results obtained are excellent. The best strategy for classification, using committee machines with descriptors of Gabor wavelets/color and volumetric fractal dimension/color, yielded a high probability of success, 96:32% in 40 classes studied. This result demonstrates how computational methods of images analysis, in particular texture analysis, can contribute and make more easier and faster the task of identifying plant species

This master thesis deals with design and implementation of algorithms for image analysis of object surface texture for the purpose of automating the surface grinding process. In the first part of this thesis, a search was performed in the field of image analysis of object surface texture. The proposed descriptors were tested on the created annotated database of texture images. Subsequently, a scene for image acquisition of the machined object was designed and assembled, and the grinding process was automated based on the results of the previous analysis. The implementation and achieved results were evaluated and other possible improvements were proposed.

This thesis deals with detection of the retinal nerve fiber layer in gray level retinal images taken by fundus camera. The first part describes a physiology of human eye and glaucoma disease. Then, the use of wavelet transform and algorithm of texture analysis applied for texture analysis. Next chapters describe theory of texture analysis named „Gray level run length matrices“ and its application for detection of the nerve fiber layer. Applications of this method are described for three types of retinal tissues and for whole image. The last chapter describes gray levels around optic disc and results obtained from parameters from GLRL matrices.

L'évolution de la texture ou de la forme d'une tumeur à l'imagerie médicale reflète les modifications internes dues à la progression (naturelle ou sous traitement) d'une lésion tumorale. Dans ces travaux nous avons souhaité étudier l'apport des caractéristiques delta-radiomiques pour prédire l'évolution de la maladie. Nous cherchons à fournir un pipeline complet de la reconstruction des lésions à la prédiction, en utilisant seulement les données obtenues en routine clinique.Tout d'abord, nous avons étudié un sous ensemble de marqueurs radiomiques calculés sur IRM, en cherchant à établir quelles conditions sont nécessaires pour assurer leur robustesse. Des jeux de données artificiels et cliniques nous permettent d'évaluer l'impact de la reconstruction 3D des zones d'intérêt et celui du traitement de l'image.Une première analyse d'un cas clinique met en évidence des descripteurs de texture statistiquement associés à la survie sans évènement de patients atteints d'un carcinome du canal anal dès le diagnostic.Dans un second temps, nous avons développé des modèles d'apprentissage statistique. Une seconde étude clinique révèle qu'une signature radiomique IRM en T2 à trois paramètres apprise par un modèle de forêts aléatoires donne des résultats prometteurs pour prédire la réponse histologique des sarcomes des tissus mous à la chimiothérapie néoadjuvante.Le pipeline d'apprentissage est ensuite testé sur un jeu de données de taille moyenne sans images, dans le but cette fois de prédire la rechute métastatique à court terme de patientes atteinte d'un cancer du sein. La classification des patientes est ensuite comparée à la prédiction du temps de rechute fournie par un modèle mécanistique de l'évolution des lésions.Enfin nous discutons de l'apport des techniques plus avancées de l'apprentissage statistique pour étendre l'automatisation de notre chaîne de traitement (segmentation automatique des tumeurs, analyse quantitative de l'oedème péri-tumoral). / Tumor shape and texture evolution may highlight internal modifications resulting from the progression of cancer. In this work, we want to study the contribution of delta-radiomics features to cancer-evolution prediction. Our goal is to provide a complete pipeline from the 3D reconstruction of the volume of interest to the prediction of its evolution, using routinely acquired data only.To this end, we first analyse a subset of MRI(-extracted) radiomics biomarquers in order to determine conditions that ensure their robustness. Then, we determine the prerequisites of features reliability and explore the impact of both reconstruction and image processing (rescaling, grey-level normalization). A first clinical study emphasizes some statistically-relevant MRI radiomics features associated with event-free survival in anal carcinoma.We then develop machine-learning models to improve our results.Radiomics and machine learning approaches were then combined in a study on high grade soft tissu sarcoma (STS). Combining Radiomics and machine-learning approaches in a study on high-grade soft tissue sarcoma, we find out that a T2-MRI delta-radiomic signature with only three features is enough to construct a classifier able to predict the STS histological response to neoadjuvant chemotherapy. Our ML pipeline is then trained and tested on a middle-size clinical dataset in order to predict early metastatic relapse of patients with breast cancer. This classification model is then compared to the relapsing time predicted by the mechanistic model.Finally we discuss the contribution of deep-learning techniques to extend our pipeline with tumor automatic segmentation or edema detection.

Corneal diseases are one of the major causes of visual impairment and blindness worldwide. Used for diagnoses, a laser confocal microscope provides a sequence of images, at incremental depths, of the various corneal layers and structures. From these, ophthalmologists can extract clinical information on the state of health of a patient’s cornea. However, many factors impede ophthalmologists in forming diagnoses starting with the large number and variable quality of the individual images (blurring, non-uniform illumination within images, variable illumination between images and noise), and there are also difficulties posed for automatic processing caused by eye movements in both lateral and axial directions during the scanning process. Aiding ophthalmologists working with long sequences of corneal image requires the development of new algorithms which enhance, correctly order and register the corneal images within a sequence. The novel algorithms devised for this purpose and presented in this thesis are divided into four main categories. The first is enhancement to reduce the problems within individual images. The second is automatic image classification to identify which part of the cornea each image belongs to, when they may not be in the correct sequence. The third is automatic reordering of the images to place the images in the right sequence. The fourth is automatic registration of the images with each other. A flexible application called CORNEASYS has been developed and implemented using MATLAB and the C language to provide and run all the algorithms and methods presented in this thesis. CORNEASYS offers users a collection of all the proposed approaches and algorithms in this thesis in one platform package. CORNEASYS also provides a facility to help the research team and Ophthalmologists, who are in discussions to determine future system requirements which meet clinicians’ needs. / The data and image files accompanying this thesis are not available online.

Dental microwear texture (DMT) analysis is used to differentiate abrasive dental wear patterns in many species fed different diets. Because DMT parameters all describe the same surface, they are expected to correlate with each other distinctively. Here, we explore the data range of, and correlations between, DMT parameters to increase the understanding of how this group of proxies records wear within and across species. The analysis was based on subsets of previously published DMT analyses in guinea pigs, sheep, and rabbits fed either a natural whole plant diet (lucerne, grass, bamboo) or pelleted diets with or without added quartz abrasives (guinea pigs and rabbits: up to 45 days, sheep: 17 months). The normalized DMT parameter range (P4: 0.69 0.25; M2: 0.83 0.16) and correlation coefficients (P4: 0.50 0.31; M2: 0.63 0.31) increased along the tooth row in guinea pigs, suggesting that strong correlations may be partially explained by data range. A comparison between sheep and guinea pigs revealed a higher DMT data range in sheep (0.93 0.16; guinea pigs: 0.47 0.29), but this did not translate into more substantial correlation coefficients (sheep: 0.35 0.28; guinea pigs: 0.55 0.32). Adding rabbits to an interspecies comparison of low abrasive dental wear (pelleted lucerne diet), the softer enamel of the hypselodont species showed a smaller data range for DMT parameters (guinea pigs 0.49 0.32, rabbit 0.19 0.18, sheep 0.78 0.22) but again slightly higher correlations coefficients compared to the hypsodont teeth (guinea pigs 0.55 0.31, rabbits 0.56 0.30, sheep 0.42 0.27). The findings suggest that the softer enamel of fast-replaced ever-growing hypselodont cheek teeth shows a greater inherent wear trace consistency, whereas the harder enamel of permanent and non-replaced enamel of hypsodont ruminant teeth records less coherent wear patterns. Because consistent diets were used across taxa, this effect cannot be ascribed to the random overwriting of individual wear traces on the more durable hypsodont teeth. This matches literature reports on reduced DMT pattern consistency on harder materials; possibly, individual wear events become more random in nature on harder material. Given the species-specific differences in enamel characteristics, the findings suggest a certain species-specificity of DMT patterns.

<p dir="ltr">Cystic fibrosis is a genetic disease that affects over 162,428 people worldwide. Currently, assessing cystic fibrosis from medical images requires a trained expert to manually annotate regions in the patient's lungs to determine the stage and severity of the disease. This process takes a substantial amount of time and effort to achieve an accurate assessment. </p><p dir="ltr">Recent advancements in machine learning and deep learning have been effective in solving classification, decision-making, identification, and segmentation problems in various disciplines. In medical research, these techniques have been used to perform image analyses that aid in organ identification, tissue classification, and lesion segmentation, which reduces the time required for physicians to analyze medical images. However, these techniques have yet to be widely applied in the assessment of cystic fibrosis. </p><p dir="ltr">This thesis describes an automated framework employed to assess the severity and extent of cystic fibrosis. The framework comprises three analysis stages: airways analysis, texture analysis, and lung lesions detection, that are utilized to extract cystic fibrosis features from CT scans, and which are used to assess the severity and extent of cystic fibrosis. The framework achieved an accuracy of 86.96\% in the staging process. The main contribution of this work is the development of a data-driven methodology used to design a quantitative cystic fibrosis staging and grading model.</p>

Thesis (MScEng)-- Stellenbosch University, 2013. / ENGLISH ABSTRACT: The measurement of key process quality variables is important for the efficient and economical operation of many chemical and mineral processing systems, as these variables can be used in process monitoring and control systems to identify and maintain optimal process conditions. However, in many engineering processes the key quality variables cannot be measured directly with standard sensors. Inferential sensing is the real-time prediction of such variables from other, measurable process variables through some form of model. In vision-based inferential sensing, visual process data in the form of images or video frames are used as input variables to the inferential sensor. This is a suitable approach when the desired process quality variable is correlated with the visual appearance of the process. The inferential sensor model is then based on analysis of the image data. Texture feature extraction is an image analysis approach by which the texture or spatial organisation of pixels in an image can be described. Two texture feature extraction methods, namely the use of grey-level co-occurrence matrices (GLCMs) and wavelet analysis, have predominated in applications of texture analysis to engineering processes. While these two baseline methods are still widely considered to be the best available texture analysis methods, several newer and more advanced methods have since been developed, which have properties that should theoretically provide these methods with some advantages over the baseline methods. Specifically, three advanced texture analysis methods have received much attention in recent machine vision literature, but have not yet been applied extensively to process engineering applications: steerable pyramids, textons and local binary patterns (LBPs). The purpose of this study was to compare the use of advanced image texture analysis methods to baseline texture analysis methods for the prediction of key process quality variables in specific process engineering applications. Three case studies, in which texture is thought to play an important role, were considered: (i) the prediction of platinum grade classes from images of platinum flotation froths, (ii) the prediction of fines fraction classes from images of coal particles on a conveyor belt, and (iii) the prediction of mean particle size classes from images of hydrocyclone underflows. Each of the five texture feature sets were used as inputs to two different classifiers (K-nearest neighbours and discriminant analysis) to predict the output variable classes for each of the three case studies mentioned above. The quality of the features extracted with each method was assessed in a structured manner, based their classification performances after the optimisation of the hyperparameters associated with each method. In the platinum froth flotation case study, steerable pyramids and LBPs significantly outperformed the GLCM, wavelet and texton methods. In the case study of coal fines fractions, the GLCM method was significantly outperformed by all four other methods. Finally, in the hydrocyclone underflow case study, steerable pyramids and LBPs significantly outperformed GLCM and wavelet methods, while the result for textons was inconclusive. Considering all of these results together, the overall conclusion was drawn that two of the three advanced texture feature extraction methods, namely steerable pyramids and LBPs, can extract feature sets of superior quality, when compared to the baseline GLCM and wavelet methods in these three case studies. The application of steerable pyramids and LBPs to further image analysis data sets is therefore recommended as a viable alternative to the traditional GLCM and wavelet texture analysis methods. / AFRIKAANSE OPSOMMING: Die meting van sleutelproseskwaliteitsveranderlikes is belangrik vir die doeltreffende en ekono-miese werking van baie chemiese– en mineraalprosesseringsisteme, aangesien hierdie verander-likes gebruik kan word in prosesmonitering– en beheerstelsels om die optimale prosestoestande te identifiseer en te handhaaf. In baie ingenieursprosesse kan die sleutelproseskwaliteits-veranderlikes egter nie direk met standaard sensors gemeet word nie. Inferensiële waarneming is die intydse voorspelling van sulke veranderlikes vanaf ander, meetbare prosesveranderlikes deur van ‘n model gebruik te maak. In beeldgebaseerde inferensiële waarneming word visuele prosesdata, in die vorm van beelde of videogrepe, gebruik as insetveranderlikes vir die inferensiële sensor. Hierdie is ‘n gepaste benadering wanneer die verlangde proseskwaliteitsveranderlike met die visuele voorkoms van die proses gekorreleer is. Die inferensiële sensormodel word dan gebaseer op die analise van die beelddata. Tekstuurkenmerkekstraksie is ‘n beeldanalisebenadering waarmee die tekstuur of ruimtelike organisering van die beeldelemente beskryf kan word. Twee tekstuurkenmerkekstraksiemetodes, naamlik die gebruik van grysskaalmede-aanwesigheidsmatrikse (GSMMs) en golfie-analise, is sterk verteenwoordig in ingenieursprosestoepassings van tekstuuranalise. Alhoewel hierdie twee grondlynmetodes steeds algemeen as die beste beskikbare tekstuuranalisemetodes beskou word, is daar sedertdien verskeie nuwer en meer gevorderde metodes ontwikkel, wat beskik oor eienskappe wat teoreties voordele vir hierdie metodes teenoor die grondlynmetodes behoort te verskaf. Meer spesifiek is daar drie gevorderde tekstuuranalisemetodes wat baie aandag in onlangse masjienvisieliteratuur geniet het, maar wat nog nie baie op ingenieursprosesse toegepas is nie: stuurbare piramiedes, tekstons en lokale binêre patrone (LBPs). Die doel van hierdie studie was om die gebruik van gevorderde tekstuuranalisemetodes te vergelyk met grondlyntekstuuranaliesemetodes vir die voorspelling van sleutelproseskwaliteits-veranderlikes in spesifieke prosesingenieurstoepassings. Drie gevallestudies, waarin tekstuur ‘n belangrike rol behoort te speel, is ondersoek: (i) die voorspelling van platinumgraadklasse vanaf beelde van platinumflottasieskuime, (ii) die voorspelling van fynfraksieklasse vanaf beelde van steenkoolpartikels op ‘n vervoerband, en (iii) die voorspelling van gemiddelde partikelgrootteklasse vanaf beelde van hidrosikloon ondervloeie. Elk van die vyf tekstuurkenmerkstelle is as insette vir twee verskillende klassifiseerders (K-naaste bure en diskriminantanalise) gebruik om die klasse van die uitsetveranderlikes te voorspeel, vir elk van die drie gevallestudies hierbo genoem. Die kwaliteit van die kenmerke wat deur elke metode ge-ekstraheer is, is op ‘n gestruktureerde manier bepaal, gebaseer op hul klassifikasieprestasie na die optimering van die hiperparameters wat verbonde is aan elke metode. In die platinumskuimflottasiegevallestudie het stuurbare piramiedes en LBPs betekenisvol beter as die GSMM–, golfie– en tekstonmetodes presteer. In die steenkoolfynfraksiegevallestudie het die GSMM-metode betekenisvol slegter as al vier ander metodes presteer. Laastens, in die hidrosikloon ondervloeigevallestudie het stuurbare piramiedes en LBPs betekenisvol beter as die GSMM– en golfiemetodes presteer, terwyl die resultaat vir tekstons nie beslissend was nie. Deur al hierdie resultate gesamentlik te beskou, is die oorkoepelende gevolgtrekking gemaak dat twee van die drie gevorderde tekstuurkenmerkekstraksiemetodes, naamlik stuurbare piramiedes en LBPs, hoër kwaliteit kenmerkstelle kan ekstraheer in vergelyking met die GSMM– en golfiemetodes, vir hierdie drie gevallestudies. Die toepassing van stuurbare piramiedes en LBPs op verdere beeldanalise-datastelle word dus aanbeveel as ‘n lewensvatbare alternatief tot die tradisionele GSMM– en golfietekstuuranalisemetodes.

Corneal diseases are one of the major causes of visual impairment and blindness worldwide. Used for diagnoses, a laser confocal microscope provides a sequence of images, at incremental depths, of the various corneal layers and structures. From these, ophthalmologists can extract clinical information on the state of health of a patient’s cornea. However, many factors impede ophthalmologists in forming diagnoses starting with the large number and variable quality of the individual images (blurring, non-uniform illumination within images, variable illumination between images and noise), and there are also difficulties posed for automatic processing caused by eye movements in both lateral and axial directions during the scanning process. Aiding ophthalmologists working with long sequences of corneal image requires the development of new algorithms which enhance, correctly order and register the corneal images within a sequence. The novel algorithms devised for this purpose and presented in this thesis are divided into four main categories. The first is enhancement to reduce the problems within individual images. The second is automatic image classification to identify which part of the cornea each image belongs to, when they may not be in the correct sequence. The third is automatic reordering of the images to place the images in the right sequence. The fourth is automatic registration of the images with each other. A flexible application called CORNEASYS has been developed and implemented using MATLAB and the C language to provide and run all the algorithms and methods presented in this thesis. CORNEASYS offers users a collection of all the proposed approaches and algorithms in this thesis in one platform package. CORNEASYS also provides a facility to help the research team and Ophthalmologists, who are in discussions to determine future system requirements which meet clinicians’ needs.