Abstract: A method for automatic segmentation of brain tumors merging full convolution neural networks with conditional random fields. The present application intends to address the issue that presently the technology of deep learning is unable to ensure the continuity of the segmentation result in shape and in space when segmenting brain tumors. For this purpose, the present application includes the following steps: step 1, processing a magnetic resonance image comprising brain tumors by utilizing a method for non-uniformity bias correction and brightness regularization, to generate a second magnetic resonance image; step 2, performing brain tumor segmentation for said second magnetic resonance image by utilizing a neural network merging a full convolutional neural network with a conditional random field, and outputting a result of brain tumor segmentation. The present method may execute brain tumor segmentation end-to-end and slice by slice during testing, which has relatively high computation efficiency.

Abstract: Various embodiments herein each include at least one of systems, devices, software, and methods for invalidation recognition and fraud prevention. One embodiment, in the form of a method, includes identifying a document-type of a received image as a check and identifying each of at least one area of the received image as a zone of the check. This method continues by processing each zone to locate text or symbols indicating the check is non-negotiable and subsequently outputting a non-negotiable rejection indication to prevent further payment processing of the check image.

Abstract: A system and method for determining a dimension of a target. The method includes: determining a camera parameter, the camera parameter including at least one of a focal length, a yaw angle, a roll angle, a pitch angle, or a height of one or more cameras; acquiring a first image and a second image of an target captured by the one or more cameras; generating a first corrected image and a second corrected image by correcting the first image and the second image; determining a parallax between a pixel in the first corrected image and a corresponding pixel in the second corrected image; determining an outline of the target; and determining a dimension of the target based at least in part on the camera parameter, the parallax, and the outline of the target.

Abstract: Systems and techniques are disclosed for improvement of machine learning systems based on enhanced training data. An example method includes providing a visual concurrent display of a set of images of features, the features requiring classification by a reviewing user. The user interface is provided to enable the reviewing user to assign classifications to the images, the user interface being configured to create, read, update, and/or delete classifications. The user interface is responsive to the user, with the user response indicating at least two images with a single classification. The user interface is updated to represent the single classification.

Abstract: The present invention relates to the field of medical monitoring, and in particular non-contact, video-based monitoring of pulse rate, respiration rate, motion, and oxygen saturation. Systems and methods are described for capturing images of a patient, producing intensity signals from the images, filtering those signals to focus on a physiologic component, and measuring a vital sign from the filtered signals. Examples include flood fill methods and skin tone filtering methods.

Abstract: An image processing apparatus includes at least one processor operatively coupled to a memory, serving as an obtaining unit configured to obtain a contrast material-enhanced image of an object, a first region extraction unit configured to extract a first region representing a first anatomical portion of the object from the image, with the first anatomical portion in the object being contrast material-enhanced, and an estimation unit configured to estimate a phase of the image based on a comparison result between a feature amount concerning a gray level in the first region and statistical data concerning a gray level in a plurality of phases of the first anatomical portion.

Abstract: Methods, systems, and devices described herein enable single-image optical sectioning, even at depth within turbid media, such as human skin or other tissue. Embodiments can eliminate the need for multiple image samples or raster scans, making in-vivo or live biological imaging easier and faster than multi-image sectioning techniques. Better contrast and resolution than traditional three-phase structured illumination microscopy (SIM) is possible in turbid media. Embodiments enable imaging of cell nuclei. Resolution and contrast resulting from disclosed embodiments are less sensitive to motion of or within patients or other targets than confocal microscopy and three-phase SIM techniques. Three-dimensional images of target specimens can be provided based on a group of single-image optical sections. Real-time imaging can also be provided.

Abstract: An encoder includes a processor and a memory. The encoder may perform a method of progressive compression. In one example implementation, the method may include identifying a pair of partitioning vertices to be connected to a split vertex associated with a collapse of an edge, creating the split vertex by collapsing the edge, encoding partitioning vertex information associated with the pair of partitioning vertices, the encoding of the partitioning vertex information based on an ordering of vertices of an umbrella of the split vertex, and the ordering of vertices of the umbrella determined based on a geometric shape and connectivity of the umbrella, and generating vertex split information that includes the partitioning vertex information. In another example implementation, the method may include entropy encoding the vertex split information prior to being transmitted.

Abstract: Soundfield signals such as e.g. Ambisonics carry a representation of a desired sound field. The Ambisonics format is based on spherical harmonic decomposition of the soundfield, and Higher Order Ambisonics (HOA) uses spherical harmonics of at least 2nd order. However, commonly used loudspeaker setups are irregular and lead to problems in decoder design. A method for improved decoding an audio soundfield representation for audio playback comprises calculating a panning function (W) using a geometrical method based on the positions of a plurality of loudspeakers and a plurality of source directions, calculating a mode matrix (?) from the loudspeaker positions, calculating a pseudo-inverse mode matrix (?+) and decoding the audio soundfield representation. The decoding is based on a decode matrix (D) that is obtained from the panning function (W) and the pseudo-inverse mode matrix (?+).

Abstract: A frame rendering method used in a head-mounted device that includes the steps outlined below. Input frames are received. First and second orientation information corresponding to a first and a second input frames are retrieved from a motion sensor. Predicted orientation information of a predicted orientation corresponding to a target time spot is generated according to the first and the second orientation information. Orientation calibration is performed on the first and the second input frames according to the first and the second orientation information to respectively generate a first and a second calibrated frames corresponding to the predicted orientation. One of a plurality of extrapolated frames corresponding to the target time spot is generated according to the first calibrated frame and the second calibrated frame.

Abstract: A three-dimensional modeling method and a three-dimensional modeling system are provided.

Abstract: A system and method for vehicle control that includes receiving an image from an imaging system. The system and method also includes classifying the traffic indicator having the color portion based on executing “You Only Look Once” (YOLO) object detection of the image and localizing the traffic indicator having the color portion with respect to a vehicle based on executing the YOLO object detection of the image. The system and method additionally includes performing “Binary Large Object” (blob) analysis based on red color components and green color components of the image to determine a color of the color portion of the traffic indicator. The system and method further includes controlling a vehicle system of the vehicle based on the classification of the traffic indicator, the location of the traffic indicator with respect to the vehicle, and the color of the color portion of the traffic indicator.

Abstract: Generation of parameterized avatars is described. An avatar generation system uses a trained machine-learning model to generate a parameterized avatar, from which digital visual content (e.g., images, videos, augmented and/or virtual reality (AR/VR) content) can be generated. The machine-learning model is trained to identify cartoon features of a particular style—from a library of these cartoon features—that correspond to features of a person depicted in a digital photograph. The parameterized avatar is data (e.g., a feature vector) that indicates the cartoon features identified from the library by the trained machine-learning model for the depicted person. This parameterization enables the avatar to be animated. The parameterization also enables the avatar generation system to generate avatars in non-photorealistic (relatively cartoony) styles such that, despite the style, the avatars preserve identities and expressions of persons depicted in input digital photographs.

Abstract: Systems and techniques are disclosed for improvement of machine learning systems based on enhanced training data. An example method includes generating an interactive classification user interface concurrently displaying a first group of medical images and a second group of medical images, each group depicting objects associated with a respective classification. User input indicating movement of medical images from the first group to the second group is detected. The moved medical images are classified according to the second group. The re-classified medical images are provided to a machine learning system, with the machine learning system updating based on analysis of object characteristics of the re-classified medical images to increase accuracies associated with automated assignment of classifications.

Abstract: Disclosed is sampling HF-QRS signals from a number of subjects (or derived values or features), and using e.g. deep learning-convolutional neural networks to find features or values which are (i) sufficiently similar for the same subject over all samples, yet (ii) sufficiently different among different subjects to allow identification. Also disclosed is finding signatures which are sufficiently stable over a particular period such that these signatures are within a deviation threshold, and then monitoring all subjects to be identified at least as often as the period used to establish the deviation threshold.

Abstract: A method includes receiving a first set of images from an image capture device of a vehicle. The method also includes performing a first analysis of movement of biomechanical points of occupants of the vehicle in the first set of images. The method further includes receiving an indication that a traffic incident has occurred. The method also includes receiving a second set of images from the image capture device corresponding to when the traffic incident occurred. The method further includes performing a second analysis of movement of the biomechanical points of the occupants in the second set of images. The method also includes determining a likelihood of injury or a severity of injury to the occupants based on the first analysis of movement and the second analysis of movement.

Abstract: The present disclosure provides a method comprising: acquiring a plurality of images of a plurality of scenes in advance, and performing feature extraction on the plurality of images respectively, to obtain a corresponding plurality of feature point sets; performing pairwise feature matching on the plurality of images, generating a corresponding eigen matrix according to the pairwise feature matching, and performing noise processing on the eigen matrix; performing 3D reconstruction according to the feature matching and the noise-processed eigen matrix and based on a ray model, to generate a 3D feature point cloud and a reconstructed camera pose set; acquiring a query image, and performing feature extraction on the query image to obtain a corresponding 2D feature point set; and performing image positioning according to the 2D feature point set, the 3D feature point cloud and the reconstructed camera pose set and based on a positioning attitude image optimization framework.

Abstract: The invention relates to a method and a device for processing a volumetric image record. The method comprises the following steps: carrying out a non-optical image scanning method on an object to be analysed and generating a volumetric image record and extracting the object contour from the volumetric image record in order to determine the position of the object surface; defining an object surface point and a surrounding area for said object surface point and analysing the grey tones within the surrounding area; calculating a quality value, which reflects the localised quality of the surface, for the object surface point on the basis of the grey-tone analysis. The device comprises equipment for carrying out the method.

Abstract: Various embodiments set forth techniques for directional sound modification. In one aspect, a system includes a plurality of audio sensors configured to acquire sound from an environment and at least one processor coupled to the plurality of audio sensors. The at least one processor is configured to determine a direction within the environment, generate an audio signal based on sound acquired from the direction within the environment, and transmit the audio signal to at least one audio output device to generate audio output. The audio output combines with sound from the direction within the environment to produce a modified sound.

Abstract: Edges of layers are detected from an input image to create a boundary line candidate image that represents the detected edges. A luminance value of the input image is differentiated to create a luminance value-differentiated image that represents luminance gradient of the layers. An evaluation score image is created which is obtained by weighting calculation at an optimum ratio between a boundary line position probability image and the luminance value-differentiated image. The boundary line position probability image is obtained from the boundary line candidate image and an existence probability image that represents existence of a boundary line to be extracted. A route having the highest total evaluation score is extracted as the boundary line. According to such an image processing apparatus and image processing method, boundary lines of layers can be extracted with a high degree of accuracy from a captured image of a target object composed of a plurality of layers.