Abstract:
Provided is a method of authenticating a user by correlating speech and corresponding lip shape. An audiovisual of a user requesting authentication is captured. The audiovisual is processed to generate a speech vector quantization sequence and a corresponding lip vector quantization sequence of the user. A likelihood of the speech vector quantization sequence and the corresponding lip vector quantization sequence with probability distributions of speech vector quantization code words corresponding to different lip shape vector quantization code words of the user requesting authentication weighed by probabilities of speech and lip vector quantization indices of the user requesting authentication is evaluated. If upon evaluation, a likelihood of the user requesting authentication being an authentic user is more than a predefined threshold, the user is authenticated.
Abstract:
The invention provides methods for detecting features of interest in cardiovascular images by receiving information from a first modality and transforming information from the first modality into a first coordinate space, receiving information from a second modality and transforming information from the second modality into a second coordinate space. The first coordinate space is aligned to the second coordinate space to combine information from the first modality and the second modality into a combined data set. The method can also involve detecting the feature of interest in a vascular image based on the combined data set.
Abstract:
Procédé de classification d'un objet de test multimodal décrit selon au moins une première et une deuxième modalités, comprenant une étape de construction hors-ligne par classification d'un dictionnaire multimédia (W m ), défini par une pluralité K m de mots multimédia, à partir d'une matrice de recodage (X) des représentants de la première modalité formant un dictionnaire de la première modalité comprenant une pluralité K T de mots de la première modalité, la matrice de recodage (X) étant construite de manière à exprimer la fréquence de chaque mot de la deuxième modalité d'un dictionnaire de la deuxième modalité comprenant une pluralité K V de mots de la deuxième modalité, pour chaque mot de la première modalité, la classification d'un objet multimodal de test (133, 533) étant réalisée en ligne au moyen d'une étape de recodage (413) de chaque représentant de la première modalité relatif à l'objet multimédia considéré sur la base du dictionnaire multimédia (W m ), suivie d'une étape d'agrégation (415) des représentants de la première modalité codés à l'étape de recodage en un unique vecteur (BoMW) représentatif de l'objet multimodal considéré.
Abstract:
Method for the automatic analysis of an image (1, 11, 12, 13) of a biological sample with respect to a pathological relevance, wherein fj local features of the image (1, 1.1, 12, 13) are aggregated to a global feature of the image (1, 11, 12, 13) using a bag of visual word approach, g) step a) is repeated at least two times using different methods resulting in at least two bag of word feature datasets,, h) computation of at least two similarity measures using the bag of word features obtained from a training image dataset and bag of word features from the image (1, 1 1, 1 2, 13) i) the image training dataset comprising a set of visual words, classifier parameters, including kernel weights and bag of word features from the training images, j) the computation of the at least two similarity measures is subject: to an adaptive computation of kernel normalization parameters and / or kernel width parameters, f) for each image (1, 11, 12, 13) one score is computed depending on the classifier parameters and kernel weights and the at least two similarity measures, the at least one score being a measure of the certainty of one pathological category compared to the image training dataset, g) for each pixel of the image (1, 11, 12, 13) a pixel-wise score is computed using the classifier parameters, the kernel weights, the at least two similarity measures, the bag of word features of the image (1, 11, 12, 13), all the local features used in the computation of the bag of word features of the image (1, 11, 12, 13) and the pixels used in the computations of the local features, h) the pixel-wise score is stored as a heatmap dataset linking the pixels of the image (1, 11, 12, 13) to the pixel-wise scores.
Abstract:
A decision-support system and computer implemented method automatically measures fee midline shift in a patient's brain using Computed Tomography (CT) images. The decision-support system and computer implemented method applies machine learning methods to features extracted from multiple sources, including midline shift, blood amount, texture pattern and other injury data, to provide a physician an estimate of intracranial pressure (ICP) levels. A hierarchical segmentation method, based on Gaussian Mixture Mode! (GMM), is used. In this approach, first an Magnetic Resonance Image (MRl) ventricle template, as prior knowledge, is used to estimate the region for each ventricle. Then, by matching the ventricle shape in CT images to fee MRl ventricle template set, the corresponding MRl slice is selected. Prom the shape matching result, the feature points for midline estimation in CT slices, such as the center edge points of the lateral ventricles, are detected. The amount of shift, along with other information such as brain tissue texture features, volume of blood accumulated in the brain, patient demographics, injury information, and features extracted from physiological signals, are used to train a machine learning method to predict a variety of important clinical factors, such as intracranial pressure (ICP), likelihood of success a particular treatment, and the need and/or dosage of particular drugs.
Abstract:
A basic idea of the present invention is to selectively employ one of at least two different feature extraction processes when generating a biometric template of an individual. An individual offers a physiological property, such as a fingerprint, an iris, an ear, a face, etc., from which biometric data can be derived, to a sensor of an enrolment authority. In the following, the property to be discussed will be fingerprints, even though any suitable biometric property may be used. From the fingerprint, a positional reference point of the biometric data is derived. The derivation of the positional reference point may be accomplished using any appropriate method out of a number of known methods. Such a reference point could be the location of a core, a delta, a weighted average of minutiae coordinates, or alike. Typically, the reference point includes a core of a fingerprint expressed as a three-dimensional coordinate denoted by means of xr, yr, and angle ar. A contribution indicator is calculated for the derived positional reference point, and it is determined whether the derived positional reference point can be considered reliable. Depending on the reliability of the derived reference point, one of the two different feature extraction processes is selected; either the first feature set is extracted using a method which is invariant of the derived reference point, or a method is used taking into account the derived reference point. The better the estimation of the reference point is, the more reliable the reference point- dependent extraction method is. Finally, the biometric template is generated using the extracted first feature set.
Abstract:
A decision-support system and computer implemented method automatically measures fee midline shift in a patient's brain using Computed Tomography (CT) images. The decision-support system and computer implemented method applies machine learning methods to features extracted from multiple sources, including midline shift, blood amount, texture pattern and other injury data, to provide a physician an estimate of intracranial pressure (ICP) levels. A hierarchical segmentation method, based on Gaussian Mixture Mode! (GMM), is used. In this approach, first an Magnetic Resonance Image (MRl) ventricle template, as prior knowledge, is used to estimate the region for each ventricle. Then, by matching the ventricle shape in CT images to fee MRl ventricle template set, the corresponding MRl slice is selected. Prom the shape matching result, the feature points for midline estimation in CT slices, such as the center edge points of the lateral ventricles, are detected. The amount of shift, along with other information such as brain tissue texture features, volume of blood accumulated in the brain, patient demographics, injury information, and features extracted from physiological signals, are used to train a machine learning method to predict a variety of important clinical factors, such as intracranial pressure (ICP), likelihood of success a particular treatment, and the need and/or dosage of particular drugs.