Abstract: Supervised machine learning using neural networks is applied to denoising images rendered by MC path tracing. Specialization of neural networks may be achieved by using a modular design that allows reusing trained components in different networks and facilitates easy debugging and incremental building of complex structures. Specialization may also be achieved by using progressive neural networks. In some embodiments, training of a neural-network based denoiser may use importance sampling, where more challenging patches or patches including areas of particular interests within a training dataset are selected with higher probabilities than others. In some other embodiments, generative adversarial networks (GANs) may be used for training a machine-learning based denoiser as an alternative to using pre-defined loss functions.

Abstract: A depth estimation method and a depth estimation apparatus of multi-view images where the method includes: taking each image among a plurality of images in a same scenario as a current image to perform the processing of: obtaining an initial depth value of each pixel in the current image; dividing the current image into a plurality of superpixels; obtaining plane parameters of the plurality of superpixels according to a predetermined constraint condition based on the initial depth values; and generating a depth value of each pixel in the superpixels based on the plane parameters of the superpixels; wherein the predetermined constraint condition includes: a co-connection constraint, which is related to a difference between depth values of adjacent points on neighboring superpixels that do not occlude each other.

Abstract: A system and method to enhance observation resolution using continuous learning include obtaining a first image of a surface area from a first satellite, and obtaining a second image of the surface area from a second satellite. The first image has a lower spatial resolution than the second image, and temporal resolution of the first images obtained by the first satellite is higher than temporal resolution of the second images obtained by the second satellite. The method also includes determining a convolution matrix A or training a neural network, obtaining additional one or more of the first images prior to obtaining an additional one of the second images, and generating a new image from each of the one or more of the first images using the convolution matrix A or the neural network. The new image has a higher spatial resolution than the one or more of the first images.

Abstract: A system and method for inspecting a composite material structure for defects includes a) an inspection apparatus having a heating device for heating a surface of the structure, an infrared camera for receiving radiation from the surface in response to heating, a controller configured to generate thermal images from the infrared radiation, b) a training system includes an arrangement for obtaining thermal images from a known composite material sample including a plurality of heating elements positioned to apply heat to an entire surface of the sample, an infrared camera for capturing thermal images of the sample, and a processing system for recording the thermal images in a training database, and c) a computer system coupled to the training system and the inspection apparatus adapted to receive thermal images from the inspection apparatus and detect parameters of defects in the structure using the training database.

Abstract: Various techniques and tools for encoding and decoding (e.g., in a video encoder/decoder) binary information (e.g., skipped macroblock information) are described. In some embodiments, the binary information is arranged in a bit plane, and the bit plane is coded at the picture/frame layer. The encoder and decoder process the binary information and, in some embodiments, switch coding modes. For example, the encoder and decoder use normal, row-skip, column-skip, or differential modes, or other and/or additional modes. In some embodiments, the encoder and decoder define a skipped macroblock as a predicted macroblock whose motion is equal to its causally predicted motion and which has zero residual error. In some embodiments, the encoder and decoder use a raw coding mode to allow for low-latency applications.

Abstract: Embodiments of the invention provide a method for scene understanding based on a sequence of image frames. The method comprises converting each pixel of each image frame to neural spikes, and extracting features from the sequence of image frames by processing neural spikes corresponding to pixels of the sequence of image frames. The method further comprises encoding the extracted features as neural spikes, and classifying the extracted features.

Abstract: Techniques for generating 3D gaze predictions based on a deep learning system are described. In an example, the deep learning system includes a neural network. A scaled image is generated from 2D image showing a user face based on a rough distance between the user eyes and a camera that generated the 2D image. Image crops at different resolutions are generated from the scaled image and include a crop around each of the user eyes and a crop around the user face. These crops are input to the neural network. In response, the neural network outputs a distance correction and a 2D gaze vector per user eye. A corrected eye-to-camera distance is generated by correcting the rough distance based on the distance correction. A 3D gaze vector for each of the user eyes is generated based on the corresponding 2D gaze vector and the corrected distance.

Abstract: In implementations of digital image search based on arbitrary image features, a server computing device maintains an images database of digital images, and includes an image search system that receives a search input as a digital image depicting image features, and receives search criteria of depicted image features in the digital image. The image search system can then determine similar images to the received digital image based on similarity criterion corresponding to the search criteria. A trained image model of the image search system is applied to determine an image feature representation of the received digital image. A feature mask model of the image search system is applied to the image feature representation to determine a masked feature representation of the received digital image. The masked feature representation of the received digital image is compared to a masked feature representation of each respective database image to identify the similar images.

Abstract: Embodiments of an automated method of processing fingerprint images, identity information is extracted from prints typically classified as having “no identification value” because of sparse or missing minutiae by capturing ridge contour information. Bezier approximations of ridge curvature are used as Ridge Specific Markers. Control points arising from Bezier curves generate unique polygons that represent the actual curve in the fingerprint. The Bezier-based descriptors are then grouped together and compared to corresponding reference print Ridge Specific Marker data. The method makes it possible to fuse a plurality of individual latent print portions into a single descriptor of identity and use the resulting data for comparison and identification. Processing of poor quality reference prints according to the methods disclosed renders these prints useable for reference purposes.

Abstract: A liveness detection system comprises a controller, a video input, a feature recognition module, and a liveness detection module. The controller is configured to control an output device to provide randomized outputs to an entity over an interval of time. The video input is configured to receive a moving image of the entity captured by a camera over the interval of time. The feature recognition module is configured to process the moving image to detect at least one human feature of the entity. The liveness detection module is configured to compare with the randomized outputs a behaviour exhibited by the detected human feature over the interval of time to determine whether the behaviour is an expected reaction to the randomized outputs, thereby determining whether the entity is a living being.

Abstract: Techniques for generating 3D gaze predictions based on a deep learning system are described. In an example, the deep learning system includes a neural network. The neural network is trained with training images. During the training, calibration parameters are initialized and input to the neural network, and are updated through the training. Accordingly, the network parameters of the neural network are updated based in part on the calibration parameters. Upon completion of the training, the neural network is calibrated for a user. This calibration includes initializing and inputting the calibration parameters along with calibration images showing an eye of the user to the neural network. The calibration includes updating the calibration parameters without changing the network parameters by minimizing the loss function of the neural network based on the calibration images. Upon completion of the calibration, the neural network is used to generate 3D gaze information for the user.

Abstract: Quality judgment on a laser beam intensity distribution is performed by taking an observation condition of the laser beam into consideration. A machine learning device includes: a state observing means that acquires data indicating an intensity distribution of a laser beam and data indicating a condition for observing the laser beam, performed to generate the data indicating the intensity distribution as input data; a label acquisition means that acquires an evaluation value related to judgment of the quality of the laser beam as a label; and a learning means that performs supervised learning using a pair of the input data acquired by the state observing means and the label acquired by the label acquisition means as training data to construct a learning model for judging the quality of the laser beam.

Abstract: A system, method, and computer program product for producing a consistent desired set of operational parameters during use of a patterning solution that would otherwise alter an initial set of operational parameters. A variable match process is adjusted dynamically to counter changes to the performance metric to tend to maintain the performance metric at a desired predetermined specification.

Abstract: The present disclosure provides a method and apparatus for infrared thermal image contour extraction. The method includes: obtaining an infrared thermal image; and determining histograms based on grayscale image of the infrared thermal image, determining segmentation thresholds of the infrared image based on the histograms, and extracting a contour based on the segmentation thresholds. The apparatus includes an image acquisition circuit, configured to capture an infrared thermal image; and a processing circuit, configured to determine histograms based on greyscale images of the infrared thermal image, to determine segmentation thresholds of the infrared thermal image based on the histograms, and to extract a contour based on segmentation thresholds.

Abstract: According to a first aspect of the present disclosure, a fingerprint authentication system is provided, comprising: a transformation unit configured to transform a first format of a captured fingerprint into a second format of the captured fingerprint, wherein the first format defines coordinates of minutia positions and the second format defines relative positions of minutiae; and an authentication unit configured to compare the relative positions with stored reference values. According to a second aspect of the present disclosure, a corresponding fingerprint authentication method is conceived. According to a third aspect of the present disclosure, a corresponding computer program product is provided.

Abstract: A method and system for performing automated defect detection is disclosed. The system may include at least one database, an image capture device and a processor. The method may comprise providing at least one database for storing information used in processing data to detect a defect in at least one member of a plurality of members in a device. The information may include a plurality of different modes of data. The method may further comprise providing a processing unit for processing the information; receiving, by the database, updates to the information; identifying a potential defect in a first mode of data; applying, by the processing unit, analysis of a second mode of data, the analysis of the second mode of data triggered by the identifying, the second mode of data different than the first mode of data; and reporting defects based on the results of the applying.

Abstract: A system and method for determining a mood for a crowd is disclosed. In example embodiments, a method includes identifying an event that includes two or more attendees, receiving at least one indicator representing emotions of attendees, determining a numerical value for each of the indicators, and aggregating the numerical values to determine an aggregate mood of the attendees of the event.

Abstract: Quality judgment on a laser beam intensity distribution is performed by taking an observation condition of the laser beam into consideration. A machine learning device includes: a state observing means that acquires data indicating an intensity distribution of a laser beam and data indicating a condition for observing the laser beam, performed to generate the data indicating the intensity distribution as input data; a label acquisition means that acquires an evaluation value related to judgment of the quality of the laser beam as a label; and a learning means that performs supervised learning using a pair of the input data acquired by the state observing means and the label acquired by the label acquisition means as training data to construct a learning model for judging the quality of the laser beam.

Abstract: Systems, methods, and articles of manufacture to perform an operation comprising processing, by a machine learning (ML) algorithm and a ML model, a plurality of images in a first dataset, wherein the ML model was generated based on a plurality of images in a training dataset, receiving user input reviewing a respective set of tags applied to each image in the first data set as a result of the processing, identifying, based on a first confusion matrix generated based on the user input and the sets of tags applied to the images in the first data set, a first labeling error in the training dataset, determining a type of the first labeling error based on a second confusion matrix, and modifying the training dataset based on the determined type of the first labeling error.

Abstract: A coding method including dividing pixels of a chrominance component of an input image into blocks having a predetermined size; selecting one among a direct current prediction method, a vertical prediction method, a horizontal prediction method, and a hybrid prediction method according to a user's input; generating a prediction value of each pixel in a current block to be predictively coded, using at least one pixel value among pixel values in an upper reference block adjacent to the current block and in a side reference block adjacent to the current block, according to the selected prediction method; generating a differential value between the prediction value and a corresponding real pixel value in the current block; and coding the differential value and information on the selected prediction method using a predetermined coding method.