Στην εργασία μας με τίτλο «Ανάπτυξη διαδικτυακού συστήματος βάσης δεδομένων με λειτουργικότητα ανάκτησης ιατρικών εικόνων» αρχικά παραθέσαμε τις τεχνικές που έχουν αναπτυχθεί από τις αρχές της δημιουργίας του τομέα της ανάκτησης εικόνας μέχρι σήμερα. Παρότι έχουν γίνει πολλές προσπάθειες για την ανάπτυξη μεθόδων οι οποίες θα βασίζονται αποκλειστικά στο περιεχόμενο τους, έως σήμερα οι περισσότερες μηχανές αναζήτησης βασίζονται ακόμη στην ομοιότητα των εικόνων βάσει των μεταδεδομένων που τις περιγράφουν.Στη συνέχεια κάναμε μια μικρή αναφορά σε ιατρικά συστήματα ανάκτησης ιατρικών εικόνων που έχουν δημιουργηθεί μέχρι σήμερα , όπως επίσης και στα αποτελέσματα του διαγωνισμού imageCLEF, ο οποίος διεξάγεται κάθε χρόνο από το 2003. Ο διαγωνισμός έχει δύο σκέλη , την ανάκτηση βάσει περιεχομένου και την ανάκτηση βάσει κειμένου, γι’ αυτό και συμμετέχουν πολλές ομάδες που ασχολούνται με την επεξεργασία φυσικής γλώσσας. Κάθε χρόνο η δυσκολία του διαγωνισμού αυξάνεται θέτοντας νέες προκλήσεις στις συμμετέχουσες ομάδες. Σύμφωνα με τα αποτελέσματα του διαγωνισμού τα καλύτερα αποτελέσματα προκύπτουν από το συνδυασμό μεθόδων και από τις δύο κατηγορίες ανάκτησης. Το σύστημα που αναπτύξαμε χρησιμοποιεί και τις δυο παραπάνω τεχνικές. Η ανάκτηση βάσει κειμένου πραγματοποιείται χρησιμοποιώντας λέξεις κλειδιά που υπάρχουν ήδη στη βάση. Ενώ για την ανάκτηση βάσει περιεχομένου εξάγουμε δύο χαρακτηριστικά , το ιστόγραμμα χρώματος και το autocorrelogram, τα οποία τα αποθηκεύουμε ως διανύσματα στη βάση και όταν θέλουμε να κάνουμε ένα ερώτημα εξάγουμε τα ίδια χαρακτηριστικά από την εικόνα ερώτημα. Η σύγκριση των δυο διανυσμάτων γίνεται υπολογίζοντας την Ευκλείδεια απόσταση μεταξύ του διανύσματος της εικόνας ερωτήματος και όλων των άλλων εικόνων της βάσης. / In our thesis, titled "Web based database system development with functionality of medical image retrieval" we present the retrieval techniques which have been developed until today. Therefore there have been done many e orts on development of methods which will rely on image content, until today most search engines (eg Google, Yahoo!) return relevant results by using text based image retrieval. Thereafter we cited some medical image retrieval systems which have been developed until today, as well as the results of imageCLEF contest, which is carried out from 2003 and every year since then. The contest has two parts, text based image retrieval and content based image retrieval, that' s the reason why many groups participated in the contest, deal with natural language processing. Every year the di culty increased and new challenges were posed to the participants. According to the results of the contest the best systems came from the combination of the two image retrieval categories. The system we developed uses the two techniques we mentioned above. Text based image retrieval is implemented by using keywords which exist in the database. While for content based image retrieval we extract two characteristics, colour histogram and autocorrelogram, which are saved as vectors in the database and when we make a query we extract the same characteristics from the image query. To compare the images we compute the distances between the image query vector and and all the other image vectors of the database. The above methods incorporated into SIDB, which is an online database management system. The system has been developed using PHP and postgreSQL and the images which have been used are medical exams from di erent parts of the human body. The biggest part of which come from the IRMA database, which has been created at Aachen University and which was used for many years in ImageCLEF competition.

Mobile devices have advanced tremendously during the last ten years and have changed our daily life in various ways. Secure pairing of mobile devices has become a significant issue considering the huge quantity of active mobile device connections and mobile traffic. However, current commonly used file sharing mobile applications rely on servers completely that are always targeted by attackers. In this thesis work, an innovative mechanism is proposed to generate symmetric keys on both mobile devices independently from a shared movement in arbitrary pattern, which means no server needs to be involved and no data exchange needed. A secret wireless-communication channel can then be established with a particular network strategy.

Over the years, cancer has increasingly become a global health problem [12]. For successful treatment, early detection and diagnosis is critical. Radiomics is the use of CT, PET, MRI or Ultrasound imaging as input data, extracting features from image-based data, and then using machine learning for quantitative analysis and disease prediction [23, 14, 19, 1]. Feature reduction is critical as most quantitative features can have unnecessary redundant characteristics. The objective of this research is to use machine learning techniques in reducing the number of dimensions, thereby rendering the data manageable. Radiomics steps include Imaging, segmentation, feature extraction, and analysis. For this research, a large-scale CT data for Lung cancer diagnosis collected by scholars from Medical University in China is used to illustrate the dimension reduction techniques via R, SAS, and Python softwares. The proposed reduction and analysis techniques were PCA, Clustering, and Manifold-based algorithms. The results indicated the texture-based features

Context. Image searching in historical handwritten documents is a challenging problem in computer vision and pattern recognition. The amount of documents which have been digitalized is increasing each day, and the task to find occurrences of a selected sub-image in a collection of documents has special interest for historians and genealogist. Objectives. This thesis develops a technique for image searching in historical documents. Divided in three phases, first the document is segmented into sub-images according to the words on it. These sub-images are defined by a features vector with measurable attributes of its content. And based on these vectors, a clustering algorithm computes the distance between vectors to decide which images match with the selected by the user. Methods. The research methodology is experimentation. A quasi-experiment is designed based on repeated measures over a single group of data. The image processing, features selection, and clustering approach are the independent variables; whereas the accuracies measurements are the dependent variable. This design provides a measurement net based on a set of outcomes related to each other. Results. The statistical analysis is based on the F1 score to measure the accuracy of the experimental results. This test analyses the accuracy of the experiment regarding to its true positives, false positives, and false negatives detected. The average F-measure for the experiment conducted is F1 = 0.59, whereas the actual performance value of the method is matching ratio of 66.4%. Conclusions. This thesis provides a starting point in order to develop a search engine for historical document collections based on pattern recognition. The main research findings are focused in image enhancement and segmentation for degraded documents, and image matching based on features definition and cluster analysis.

Defected wheel are one the major reasons endangered state of railroad vehicles safety statue, due to vehicle derailment and worsen the quality of freight and passenger transportation. Therefore, timely defect detection for monitoring and detecting the state of defects is highly critical. This thesis presents a passive non-contact acoustic structural health monitoring approach using ultrasonic acoustic emissions (UAE) to detect certain defects on different structures, as well as, classifying the type of the defect on them. The acoustic emission signals used in this study are in the ultrasonic range (18-120 kHz), which is significantly higher than the majority of the research in this area thus far. For the proposed method, an impulse excitation, such as a hammer strike, is applied to the structure. In addition, ultrasound techniques have higher sensitivity to both surface and subsurface defects, which make the defect detection more accurate. Three structures considered for this study are: 1) a longitudinal beam, 2) a lifting weight, 3) an actual rail-wheel. A longitudinal beam was used at the first step for a better understanding of physics of the ultrasound propagation from the defect, as well, develop a method for extracting the signature response of the defect. Besides, the inherent directionality of the ultrasound microphone increases the signal to noise ratio (SNR) and could be useful in the noisy areas. Next, by considering the ultimate goal of the project, lifting weight was chosen, due to its similarity to the ultimate goal of this project that is a rail-wheel. A detection method and metric were developed by using the lifting weight and two type of synthetic defects were classified on this structure. Also, by using same extracted features, the same types of defects were detected and classified on an actual rail-wheel. / Master of Science

Esta tese apresenta uma metodologia para classificação automática de achados mamográficos em mamas densas através de uma abordagem híbrida de classificadores e extração de atributos, como parte de um esquema computadorizado de auxílio ao diagnóstico (CAD) para mamografia. Foram implementadas duas redes Backpropagation. Uma responde pela classificação de clusters de microcalcificações, através de atributos descritores geométricos, em duas classes - suspeito e não suspeito. A outra rede classifica nódulos utilizando descritores geométricos e uma entrada com informação extraída de atributos de intensidade, produzindo na saída dois tipos de informação: presença ou não do nódulo, e constatada a presença do nódulo, classificação da região de interesse (RI) entre as categorias BI-RADS. As respostas de um \"clusterizador\" de RIs através de atributos de intensidade serviram de entrada para essa rede fornecendo uma informação de grau de densidade da RI. Uma interface foi desenvolvida para a apresentação dos resultados a fim de fornecer informações mais detalhadas da classificação e do caso analisado. Os resultados do classificador foram analisados através de análise estatística de sensibilidade e especificidade, e também por curvas ROC. Os resultados obtidos ficaram próximos a 89% de acerto total (verdadeiros-positivos mais verdadeiros-negativos) para nódulos produzindo valor de Az superior a 0,92 e ultrapassaram 75% de acerto da classificação entre as classes BI-RADS. Na classificação dos clusters os acertos totais do classificador ficaram próximos de 90%, com Az superior a 0,94. Para ambos tipos de lesões, as taxas de respostas falsas-negativas ficaram abaixo de 0,1, o que significa baixo erro em relação à não detecção da doença quando o sinal está presente. O classificador apresentado nesse trabalho é a conclusão de uma etapa importante do esquema CAD que vem sendo desenvolvido pelo grupo, além de possibilitar a disponibilização de mais uma ferramenta automática de auxílio ao diagnóstico do câncer de mama aos serviços de mamografia. / This thesis presents a methodology for automatic classification of mamographic findings in image of dense breast through hybrid approach of classifiers and features extraction techniques, as part of a computer-aided diagnostic (CAD) scheme for mammography. Two Backpropagation neural networks were implemented. One for microcalcifications clustered classification, through geometric descriptors, in two classes - suspect and non-suspect. The other neural network classifies nodules using geometric descriptors and additional information from intensity features extracted, producing in the output two kinds of information: presence or not of the nodule, and if nodule is present in the image, classification among BI-RADS categories. The result of clustering technique by using intensity features is presented as a new input to neural network, supplying density degree of image. An interface was developed for results presentation in order to supply more detailed information from the classifier outputs and of the analyzed case. The results of the classifier were analyzed through sensibility and specificity statistical analysis, and also for ROC curves. The results were close to 89% of total accuracy (positive-true plus negative-true) for nodules producing value of Az more than 0,92 and 75% of accuracy to classification among BI-RADS categories. In the cluster classification the total accuracy is about 90%, and Az greater than 0,94. In both kinds of lesions, negative-false result rates were below 0,1, which means low error related to the fail to detect the disease when the signal is present. The classifier presented in this work is the conclusion of an important stage of the CAD scheme that has been developed by the group, besides making possible the availability of one more automatic tool of aid to the breast cancer diagnosis to be used in mammography centers.

This dissertation describes a new methodology for multi-modal (2-D + 3-D) face modeling and recognition. There are advantages in using each modality for face recognition. For example, the problems of pose variation and illumination condition, which cannot be resolved easily by using the 2-D data, can be handled by using the 3-D data. However, texture, which is provided by 2-D data, is an important cue that cannot be ignored. Therefore, we use both the 2-D and 3-D modalities for face recognition and fuse the results of face recognition by each modality to boost the overall performance of the system. In this dissertation, we consider two different cases for multi-modal face modeling and recognition. In the first case, the 2-D and 3-D data are registered. In this case we develop a unified graph model called Attributed Relational Graph (ARG) for face modeling and recognition. Based on the ARG model, the 2-D and 3-D data are included in a single model. The developed ARG model consists of nodes, edges, and mutual relations. The nodes of the graph correspond to the landmark points that are extracted by an improved Active Shape Model (ASM) technique. In order to extract the facial landmarks robustly, we improve the Active Shape Model technique by using the color information. Then, at each node of the graph, we calculate the response of a set of log-Gabor filters applied to the facial image texture and shape information (depth values); these features are used to model the local structure of the face at each node of the graph. The edges of the graph are defined based on Delaunay triangulation and a set of mutual relations between the sides of the triangles are defined. The mutual relations boost the final performance of the system. The results of face matching using the 2-D and 3-D attributes and the mutual relations are fused at the score level. In the second case, the 2-D and 3-D data are not registered. This lack of registration could be due to different reasons such as time lapse between the data acquisitions. Therefore, the 2-D and 3-D modalities are modeled independently. For the 3-D modality, we developed a fully automated system for 3-D face modeling and recognition based on ridge images. The problem with shape matching approaches such as Iterative Closest Points (ICP) or Hausdorff distance is the computational complexity. We model the face by 3-D binary ridge images and use them for matching. In order to match the ridge points (either using the ICP or the Hausdorff distance), we extract three facial landmark points: namely, the two inner corners of the eyes and the tip of the nose, on the face surface using the Gaussian curvature. These three points are used for initial alignment of the constructed ridge images. As a result of using ridge points, which are just a fraction of the total points on the surface of the face, the computational complexity of the matching is reduced by two orders of magnitude. For the 2-D modality, we model the face using an Attributed Relational Graph. The results of the 2-D and 3-D matching are fused at the score level. There are various techniques to fuse the 2-D and 3-D modalities. In this dissertation, we fuse the matching results at the score level to enhance the overall performance of our face recognition system. We compare the Dempster-Shafer theory of evidence and the weighted sum rule for fusion. We evaluate the performance of the above techniques for multi-modal face recognition on various databases such as Gavab range database, FRGC (Face Recognition Grand Challenge) V2.0, and the University of Miami face database.

Esta tese apresenta uma metodologia para classificação automática de achados mamográficos em mamas densas através de uma abordagem híbrida de classificadores e extração de atributos, como parte de um esquema computadorizado de auxílio ao diagnóstico (CAD) para mamografia. Foram implementadas duas redes Backpropagation. Uma responde pela classificação de clusters de microcalcificações, através de atributos descritores geométricos, em duas classes - suspeito e não suspeito. A outra rede classifica nódulos utilizando descritores geométricos e uma entrada com informação extraída de atributos de intensidade, produzindo na saída dois tipos de informação: presença ou não do nódulo, e constatada a presença do nódulo, classificação da região de interesse (RI) entre as categorias BI-RADS. As respostas de um \"clusterizador\" de RIs através de atributos de intensidade serviram de entrada para essa rede fornecendo uma informação de grau de densidade da RI. Uma interface foi desenvolvida para a apresentação dos resultados a fim de fornecer informações mais detalhadas da classificação e do caso analisado. Os resultados do classificador foram analisados através de análise estatística de sensibilidade e especificidade, e também por curvas ROC. Os resultados obtidos ficaram próximos a 89% de acerto total (verdadeiros-positivos mais verdadeiros-negativos) para nódulos produzindo valor de Az superior a 0,92 e ultrapassaram 75% de acerto da classificação entre as classes BI-RADS. Na classificação dos clusters os acertos totais do classificador ficaram próximos de 90%, com Az superior a 0,94. Para ambos tipos de lesões, as taxas de respostas falsas-negativas ficaram abaixo de 0,1, o que significa baixo erro em relação à não detecção da doença quando o sinal está presente. O classificador apresentado nesse trabalho é a conclusão de uma etapa importante do esquema CAD que vem sendo desenvolvido pelo grupo, além de possibilitar a disponibilização de mais uma ferramenta automática de auxílio ao diagnóstico do câncer de mama aos serviços de mamografia. / This thesis presents a methodology for automatic classification of mamographic findings in image of dense breast through hybrid approach of classifiers and features extraction techniques, as part of a computer-aided diagnostic (CAD) scheme for mammography. Two Backpropagation neural networks were implemented. One for microcalcifications clustered classification, through geometric descriptors, in two classes - suspect and non-suspect. The other neural network classifies nodules using geometric descriptors and additional information from intensity features extracted, producing in the output two kinds of information: presence or not of the nodule, and if nodule is present in the image, classification among BI-RADS categories. The result of clustering technique by using intensity features is presented as a new input to neural network, supplying density degree of image. An interface was developed for results presentation in order to supply more detailed information from the classifier outputs and of the analyzed case. The results of the classifier were analyzed through sensibility and specificity statistical analysis, and also for ROC curves. The results were close to 89% of total accuracy (positive-true plus negative-true) for nodules producing value of Az more than 0,92 and 75% of accuracy to classification among BI-RADS categories. In the cluster classification the total accuracy is about 90%, and Az greater than 0,94. In both kinds of lesions, negative-false result rates were below 0,1, which means low error related to the fail to detect the disease when the signal is present. The classifier presented in this work is the conclusion of an important stage of the CAD scheme that has been developed by the group, besides making possible the availability of one more automatic tool of aid to the breast cancer diagnosis to be used in mammography centers.

Cancer is a leading cause of death globally, and early detection is crucial for better outcomes. This research aims to improve Region Of Interest (ROI) segmentation and feature extraction in medical image analysis using Radiomics techniques with 3D Slicer, Pyradiomics, and Python. Dimension reduction methods, including PCA, K-means, t-SNE, ISOMAP, and Hierarchical Clustering, were applied to highdimensional features to enhance interpretability and efficiency. The study assessed the ability of the reduced feature set to predict T-staging, an essential component of the TNM system for cancer diagnosis. Multinomial logistic regression models were developed and evaluated using MSE, AIC, BIC, and Deviance Test. The dataset consisted of CT and PET-CT DICOM images from 131 lung cancer patients. Results showed that PCA identified 14 features, Hierarchical Clustering 17, t-SNE 58, and ISOMAP 40, with texture-based features being the most critical. This study highlights the potential of integrating Radiomics and unsupervised learning techniques to enhance cancer prediction from medical images.

The research work presented in this thesis introduces novel approaches for both visual region of interest extraction and visual feature extraction for use in audio-visual automatic speech recognition. In particular, the speaker‘s movement that occurs during speech is used to isolate the mouth region in video sequences and motionbased features obtained from this region are used to provide new visual features for audio-visual automatic speech recognition. The mouth region extraction approach proposed in this work is shown to give superior performance compared with existing colour-based lip segmentation methods. The new features are obtained from three separate representations of motion in the region of interest, namely the difference in luminance between successive images, block matching based motion vectors and optical flow. The new visual features are found to improve visual-only and audiovisual speech recognition performance when compared with the commonly-used appearance feature-based methods. In addition, a novel approach is proposed for visual feature extraction from either the discrete cosine transform or discrete wavelet transform representations of the mouth region of the speaker. In this work, the image transform is explored from a new viewpoint of data discrimination; in contrast to the more conventional data preservation viewpoint. The main findings of this work are that audio-visual automatic speech recognition systems using the new features extracted from the frequency bands selected according to their discriminatory abilities generally outperform those using features designed for data preservation. To establish the noise robustness of the new features proposed in this work, their performance has been studied in presence of a range of different types of noise and at various signal-to-noise ratios. In these experiments, the audio-visual automatic speech recognition systems based on the new approaches were found to give superior performance both to audio-visual systems using appearance based features and to audio-only speech recognition systems.