Abstract: Embodiments relate to characterizing a water flow. Initially, a camera targeted at the water flow obtains images. An image pyramid is generated for each image, and a first level of each image pyramid is tiled at an initial scale to identify initial features. The initial features are matched across the first level of the image pyramids at a pixel-level to obtain a first set of feature displacements. A second level of each image pyramid is tiled at a decreased scale to identify a scaled set of features. The scaled features are matched across the second level of the image pyramids at the pixel-level to obtain a second set of feature displacements. Optical flow algorithms are applied to modify each feature displacement to account for a corresponding subpixel registration. A stereo reconstruction is applied to the feature displacements to generate a three dimensional (3D) displacement map.

Abstract: An image processing device includes a processor including hardware. The processor implements an image acquisition process that acquires time-series images that were captured in time series, and a process that detects a change position, the change position being a position at which an object captured within the time-series images changes from a first object to a second object. The processor implements the process that extracts a feature quantity from each of the time-series images, determines whether the object captured within each of the time-series images is the first object or the second object based on the feature quantity, extracts candidate positions for the change position based on the determination results, and compares the candidate positions based on the determination results to determine the change position.

Abstract: Methods and systems may help to provide filtered notification content that provides useful information to the intended recipient, but does not provide the same information to an unauthorized viewer. To do so, when an application event occurs, filtered notification content may be generated. The filtered notification content may combine a non-obfuscated portion of the event content and an obfuscated portion of the event content, such that when viewed by an intended recipient, the filtered notification content provides implicit information that the device expects the particular user to understand based on the user's own experience and/or other factors.

Abstract: An image recognition accelerator, a terminal device, and an image recognition method are provided. The image recognition accelerator includes a dimensionality-reduction processing module, an NVM, and an image matching module. The dimensionality-reduction processing module first reduces a dimensionality of first image data. The NVM writes, into a first storage area of the NVM according to a specified first current I, ? low-order bits of each numeric value of the first image data on which dimensionality reduction has been performed, and writes, into a second storage area of the NVM according to a specified second current, (N??) high-order bits of each numeric value of the first image data on which dimensionality reduction has been performed. The image matching module determines whether an image library stored in the NVM includes image data matching the first image data on which dimensionality reduction has been performed.

Abstract: In order to decrease a memory footprint for volume rendering, distance field information is adaptively compressed and merged with an existing data structure having unused memory to form a merged dataset. During rendering, the merged dataset is sampled, and a sampled value is used to identify a look-up table for a corresponding voxel or is decompressed and used to skip empty space.

Abstract: Methods and systems for automatically providing a notification of an operational problem between two devices are provided. The method may include searching online content for user feedback comments associated with a first device and a second device. Sentiment analysis may be performed for the user feedback comments and an operational problem that occurs between the first device and the second device may be identified based on the sentiment analysis. Next, a notification of the operational problem may be provided. Additionally, a probable cause associated with the operational problem may be identified and a suggested solution based on the identified probable cause may be provided. The suggested solution may correct the operational problem between the first device and the second device.

Abstract: A method for fine-grained object recognition in a robotic system is disclosed that includes obtaining an image of an object from an imaging device. Based on the image, a deep category-level detection neural network is used to detect pre-defined categories of objects. A feature map is generated for each pre-defined category of object detected by the deep category-level detection neural network. Embedded features are generated, based on the feature map, using a deep instance-level detection neural network corresponding to the pre-defined category of the object, wherein each pre-defined category of an object comprises a corresponding different instance-level detection neural network. An instance-level of the object is determined based on classification of the embedded features.

Abstract: The present disclosure relates to methods and systems for storing and querying data. According to the embodiments of the present invention, two-layer indexes are created for multi-dimension data, wherein the primary index is created based on two or more dimensions to retrieve respective data units of the data, while the secondary index is created based on specific dimensions to retrieve respective data blocks in the data unit. Correspondingly, when receiving a multi-dimension query request for data, the primary retrieval first determines a data unit including the target data based on a primary index, and then the secondary retrieval quickly locates a data block including the target data based on the secondary index. In this way, the multi-dimension retrieval can be efficiently performed. Moreover, by appropriately setting the size of a smallest data block, the I/O efficiency of data access will be significantly enhanced.

Abstract: The disclosure discloses methods and systems for automatically sorting a plurality of pages of a document by a multi-function device. The device comprises a document feeder, a document swapper and a document collector. The multi-function device receives a request to sort the plurality of pages in accordance with at least one sorting parameter. The multi-function device scans the pages and identifies one or more character fields for each of the scanned pages and executes a plurality of page swap cycles between the document feeder, the document swapper, and the document collector until each of the scanned pages are sorted based on the at least one sorting parameter. The sorting is performed by comparing the character fields for each of the scanned pages, a first subset of scanned pages are placed in the document swapper and a second subset is placed in the document collector in each page swap cycle.

Abstract: There is provided a class estimation apparatus. A data reception unit receives an input of a feature amount of data. An identification device reception unit receives an input of an identification device having a tree structure. A table storage unit stores a probability table and a feature amount table in a leaf node of the tree structure. A probability table correction unit corrects the probability table. A class estimation unit estimates a class of the data.

Abstract: An image reconstruction processing method according to this invention is characterized by adding weight to conventional reconstruction processing steps. That is, weighting is performed when performing the reconstruction process based on optimization calculations of a multivariate function in which a digital image is unknown, and which consists of data function generalized from the likelihood function of Poisson distribution. When a weight coefficient is a constant independent of element data, for example, the weight coefficient is set based on directionality of linear noises occurring in a reconstruction image, or the weight coefficient is set based on detection depth position information of detector elements of DOI detectors (step S1). By weighting partial functions with weight coefficients of back projection to a reconstruction image of the element data corresponding to the partial functions, artifacts appearing on the image can be suppressed, or the spatial resolution of the image can be improved.

Abstract: A method and system are provided for targeting specific users with specific requested actions to be performed by the user and verified by any methods on behalf of an advertiser within an Internet display advertising campaign. The system discloses techniques for receiving an advertiser's requested actions and corresponding definitions of what constitutes a satisfaction of the action. The disclosure also details techniques for determining if and when and to what degree a requested action has been completed or satisfied by the targeted user, and details corresponding techniques for compensating the user and any other real or virtual entity who had contributed to the satisfaction of the requested action. The system implements a network of sensors that can aid in the determination of if and when and to what degree a requested action has been completed or satisfied. The network of sensors registers sensor recordings within a specialized sensor recording marketplace.

Abstract: Embodiments of the present disclosure include methods, devices, and computer program products for detecting malware in a file. Embodiments include identifying a plurality of features of the file, categorizing each of the plurality of features to define a plurality of categories of features, building a first decision tree based on a first category from the plurality of categories, the first category comprising a first set of features of the file, and building a second decision tree based on a second category from the plurality of categories, the second decision tree comprising a second set of features of the file, the second set different from the first set. Some embodiments include comparing results from each decision tree to determine the presence or absence of malware.

Abstract: There is disclosed a method of generating an index of images, the index of images for enabling comparison of the image against other images, the method executable at a server. The method comprises: determining at least one key for the index, the at least one key including at least a portion of a visual features composite parameter associated with an image to be indexed, the visual features composite parameter having been determined by executing steps of; identifying a first local region of the image and a second local region of the image.

Abstract: A maintenance system can be used with a self-driving vehicle. The maintenance system can include a camera system coupled to an interior of the vehicle and a vehicle management system configured to autonomously drive the vehicle to a first location to remove an item left in the vehicle by a rider.

Abstract: In an information processing method, the importance degree of each person who appears in a first image group and the importance degree of each person who appears in a second image group, containing the first image group, are determined, and, when selecting the image of the output target from the first image group, in a case in which the importance degree of a candidate image of the output target included in the first image group, which is determined from the appearance count in the first image group, exceeds a first threshold, and, in a case in which the importance degree of the candidate image of the output target, which is determined from the appearance count in the second image group, exceeds a second threshold, the candidate image is selected as the output target.

Abstract: Systems, apparatuses, and methods for enhanced resolution video and security via machine learning are disclosed. A transmitter reduces a resolution of each image of a videostream from a first, higher image resolution to a second, lower image resolution. The transmitter generates a set of parameters for programming a neural network to reconstruct a version of each image at the first image resolution. Then, the transmitter sends the images at the second image resolution to the receiver, along with the first set of parameters. The receiver programs a neural network with the first set of parameters and uses the neural network to reconstruct versions of the images at the first image resolution. The transmitter can send the first set of parameters to the receiver via a secure channel, ensuring that only the receiver can decode the images from the second image resolution to the first image resolution.

Abstract: A pattern classifying apparatus according to the present invention accesses at least two of a plurality of memories holding data to be used in pattern classification in parallel so as to read a plurality of data pieces and performs pattern classification with weak classifiers by using the plurality of read data pieces.

Abstract: A method includes identifying, by a processor, a gauge dial pattern in a first file based on a gauge dial template in a second file; identifying, by the processor, a needle pattern in the first file; generating, by the processor, a line pattern that approximates the needle pattern; determining, by the processor, a deflection angle of the needle pattern with respect to the gauge dial pattern based on the line pattern; converting, by the processor, the deflection angle into a reading of the gauge dial pattern based on the gauge dial template; and taking, by the processor, an action based on the reading.

Abstract: A user terminal device is provided. The user terminal device includes a plurality of different types of sensors, a storage unit configured to store object information, and a controller configured to acquire sensing results of the plurality of sensors, determine a plurality of object categories respectively corresponding to the sensing results, and compare objects belonging to the plurality of object categories to estimate an object commonly belonging to the object categories as a recognition result.

Abstract: The present invention is directed at a system, method and device for detecting offensive content on a portable electronic device, by monitoring communications sent, received or stored on the portable electronic device, and wherein monitoring comprises collecting content data, classifying content data by calculating an alert score for content data wherein an alert score corresponds to offensive content detected, and sending an alert notification to a second portable electronic device to alert the detection of offensive content on the first portable electronic device.

Abstract: Provided herein are display systems and units, including those configured for dynamic communication in a physical location, such as in retail settings. Also included herein are methods for dynamically displaying product information in a physical location, such as a retail setting.

Abstract: Systems and methods disclosed herein associate images with business venues. An example method includes: receiving a first image and retrieving textual reviews and stored images that are associated with one or more candidate business venues. The method further includes: detecting, using trained visual detectors, a plurality of business-aware concepts in the first image and assessing likelihood that detected business-aware concepts are in the first image. The method additionally includes: (i) generating a first representation of the first image based on the likelihoods and one or more term vectors for high-scoring concepts and (ii) receiving second representations of each candidate based on the retrieved textual reviews and stored images. In accordance with determining that the first representation is most similar to a respective second representation of a first candidate, the method includes: (i) associating the first image with the first candidate and (ii) providing an indication of the association.

Abstract: Implementations include actions of receiving an initial data set including a plurality of images, each image being associated with a set of labels, wherein each label in the set of labels is assigned to a respective image of the plurality of images by an initial model, the initial model being specific to a ground-truth label; for each image in the plurality of images: providing a list of categories associated with a respective image based on a respective set of labels, and determining a primary category of the respective image based on the list of categories; determining a category of the ground-truth label; and providing a revised data set based on the initial data set by comparing the category to primary categories of respective images in the plurality of images, the initial model being trained based on the revised data set to provide a revised model.

Abstract: A specified object in still images or video may be detected using a sliding search window technique, applied to the original image and its downscaled versions, in order to detect objects of different sizes. At each scale and each position of the sliding window, the technique may use a boosted tree classifier to determine whether the window contains the object. It may exit earlier if some intermediate sum falls below the certain threshold. To accelerate object detection, hybrid features are used in different boosted chains. For first boosted chains, the fastest features may be applied and then, after more complex (but slower) features and for the last few chains, the most powerful feature (but most computationally expensive) is used. This strategy may improve the speed of detection because for a majority of checking windows, only first boosted chains are used and so only the fastest features are calculated in some embodiments.

Abstract: A system and method to determine a flow direction automatically so that counter-flow motion may be detected without user calibration is disclosed. The system and method utilize information available from existing video compression algorithms to sense motion and determine a flow direction. Once a flow direction for an area is determined, the system and method can detect counter-flow motion using compressed video without any decompression, thereby improving the speed of detection and simplifying the computational complexity. These improvements facilitate counter-flow alarms and expand the information available to a user monitoring the video.

Abstract: Roughly described, a method for estimating a size measurement of a subject feature on a body, for use with a photo, snapped by a camera which may be handheld. A reference marker is placed on the body near the subject feature. The reference marker includes concentric rings to help confirm its identify, and registration markings to determine the pose at which the photo was taken. The registration markings can be dots at vertices of a square. The method can include detecting concentric ellipses in the photo corresponding to the concentric circles, detecting the registration markings in the photo within a predetermined region which is outside the detected ellipses, geometrically transforming at least the subject feature in the photo based on the detected locations of the registration markings, calculating the scale of the photo based on the detected registration markings, and providing the transformed image and scale for the size measurement.

Abstract: A method for obtaining light field data of a scene from a sequence of images of the scene, each image being captured at a different viewpoint using a non-light-field image capture device is described. The method includes obtaining a reference image of the scene at a reference viewpoint, an interest area of the scene being included in the reference image and a reference position (x0,y0) being associated with the interest area in a reference local coordinate system of the capture device.

Abstract: A method for pyramid vector quantization indexing of audio/video signals comprises obtaining of an integer input vector representing the audio/video signal samples. A leading sign is extracted from the integer input vector. The leading sign is a sign of a terminal non-zero coefficient in the integer input vector. The terminal non-zero coefficient is one of a first non-zero coefficient and a last non-zero coefficient in the integer input vector. The integer input vector is indexed with a pyramid vector quantization enumeration scheme into an output index representing the audio/video signal samples. The pyramid vector quantization enumeration scheme is designed for neglecting the sign of the terminal non-zero coefficient. The output index and the leading sign are outputted. A corresponding method for de-indexing, an encoder, a decoder, and computer programs therefore are also disclosed.

Abstract: A method of navigating an autonomous coverage robot on a floor includes controlling movement of the robot across the floor in a cleaning mode, receiving a sensor signal indicative of an obstacle, rotating the robot away from the sensed obstacle, determining a change in the received sensor signal during at least a portion of the rotation of the robot away from the sensed obstacle, and identifying the sensed obstacle based at least in part on the determined change in the received sensor signal.

Abstract: Various implementations of the invention perform facial recognition on a target image compared against a color image from an image gallery, where the target image was acquired by an infrared-sensitive camera of a target that was illuminated with infrared light. According to various implementations of the invention, a blue component and a green component of the pixels in the color image are suppressed or eliminated, and facial recognition is performed between the target image and the color-suppressed image.

Abstract: Method and system for enhancing visual perception of augmented reality presentation. The location and line-of-sight (LOS) of a user wearing a see-through head-mounted display (HMD) is detected. A future background environment to be viewed by the user through the HMD is predicted based on at least the detected location and detected LOS. At least one color-attribute of the future background environment at a background location corresponding to a user LOS is predicted. The predicted color-attribute is compared with at least one color-attribute of an intended foreground supplementary image to be projected on the HMD overlaid onto the future background environment at the background location. When an incompatibility condition is detected, at least one visual parameter of the supplementary image is adjusted to minimize the incompatibility condition, and the supplementary image is projected on the HMD at the background location with the adjusted visual parameter.

Abstract: Particular embodiments of a computing device determine unoccupied positions in a display grid, wherein the grid corresponds to specifications for a displayable region of a screen associated with a computing device. A plurality of image elements are identified as candidates for placement within the grid. Then, while at least one unoccupied position and at least one candidate image element remain, one of the candidate image elements is matched to a set of the unoccupied positions, and the positions in the matched set are designated as occupied by the matched image element. Matching may be based on a pattern of positions, which is determined using the size and the shape of the image element and a size of a position in the grid. Layout information is then provided to render the image elements in accordance with placement of each matched image element at the designated positions in the grid.

Abstract: Embodiments of the invention relate to file sharing, and more specifically to sharing image files based on a social networking relationship. A digital image and associated file are created with geotag data that defines an image capture. Data associated with the image capture is stored as image metadata associated with the image file. A hyperlink to a preferred viewing URL is created and established based on a match produced from a comparison of the image file image capture data with at least one physical location identified in an index of cached data. The hyperlink is associated with the image file and directs a viewer to the physical and/or virtual location where the image was captured.

Abstract: Video tracking systems and methods include a peripheral master tracking process integrated with one or more tunnel tracking processes. The video tracking systems and methods utilize video data to detect and/or track separately several stationary or moving objects in a manner of tunnel vision. The video tracking system includes a master peripheral tracker for monitoring a scene and detecting an object, and a first tunnel tracker initiated by the master peripheral tracker, wherein the first tunnel tracker is dedicated to track one detected object.

Abstract: A method includes identifying, by a processor, a gauge dial pattern in a first file based on a gauge dial template in a second file; identifying, by the processor, a needle pattern in the first file; generating, by the processor, a line pattern that approximates the needle pattern; determining, by the processor, a deflection angle of the needle pattern with respect to the gauge dial pattern based on the line pattern; converting, by the processor, the deflection angle into a reading of the gauge dial pattern based on the gauge dial template; and taking, by the processor, an action based on the reading.

Abstract: A method for presenting interactive multimedia programs includes selecting a multimedia program having a plurality of segments, classifying the multimedia program and the plurality of segments according to content, and generating a plurality of additional content items based on the classification, each of which being associated with at least one of the plurality of segments. The method further includes generating a plurality of pre-defined queries for each of the plurality of segments based on a content of the segment and at least one speech recognition grammar based on the plurality of pre-defined queries. The plurality of additional content items are retrievable responsive to the plurality of pre-defined queries. The method further includes generating a data stream including the at least one speech recognition grammar and all of the plurality of additional content items and the plurality of pre-defined queries arranged according to an order of the plurality of segments.

Abstract: A computing apparatus is configured to operate a controller on a communication log file to infer a message structure for communications between a remote control and a controlled device. The controller applies Hidden Markov Model and Finite State Machine to the message structure to operate the remote control on the controlled device to perform predefined actions, receives a state of the controlled device, and generates a semantic classification for the message structure from the state, the semantic classification applied to operation of the controller.

Abstract: Systems, methods, and non-transitory computer-readable media can identify a content item posted by a user. Time data associated with the content item can be acquired. A profile picture can be identified, based at least in part on the time data, out of a set of profile pictures associated with the user. The profile picture can be provided in association with the content item.

Abstract: Implementations relate to providing animated user identifiers. In some implementations, a computer-executed method includes determining that a video call over a communication network is connected between a first device associated with a first user and a second device associated with a second user. The method stores a set of multiple images that are received by the first device as part of the video call, and forms a motion clip including the set of multiple images and indicating a sequence of the set of multiple images for display. The method assigns the motion clip to a user identifier associated with the second user, and causes display of the motion clip to visually represent the second user in response to the user identifier being displayed in at least one user interface on the first device.

Abstract: A method of performing an image autofocus operation using multiple cameras includes performing, at an image processor, a first autofocus operation on a first region of interest in a scene captured by a first camera and determining a second region of interest in the scene captured by a second camera. The second region of interest is determined based on the first region of interest. The method further includes performing a second autofocus operation on the second region of interest. The method also includes fusing a first image of the scene captured by the first camera with a second image of the scene captured by the second camera to generate a fused image. The first image is based on the first autofocus operation and the second image is based on the second autofocus operation.

Abstract: A method is described that includes comparing a characteristic of an image to stored characteristics of spam images. The method also includes generating a signature of the present image. The method further includes comparing the signature of the present image to stored signatures of spam images. The method also includes determining the spam features corresponding to the stored signatures of spam images that match the signature of the present image.

Abstract: The invention is directed towards segmenting images based on natural language phrases. An image and an n-gram, including a sequence of tokens, are received. An encoding of image features and a sequence of token vectors are generated. A fully convolutional neural network identifies and encodes the image features. A word embedding model generates the token vectors. A recurrent neural network (RNN) iteratively updates a segmentation map based on combinations of the image feature encoding and the token vectors. The segmentation map identifies which pixels are included in an image region referenced by the n-gram. A segmented image is generated based on the segmentation map. The RNN may be a convolutional multimodal RNN. A separate RNN, such as a long short-term memory network, may iteratively update an encoding of semantic features based on the order of tokens. The first RNN may update the segmentation map based on the semantic feature encoding.

Abstract: Embodiments provide techniques for identifying unique data elements in data record, removing the unique data elements from the record, storing the record, and employing the removed, unique data as a key in an index into the data storage. When the record is retrieved based on the key in the index, the removed unique data elements is restored. Removal of the unique data elements from the records may enable more efficient compression of the records using compression algorithms that employ an external reference dictionary. In some embodiments, a combination of multiple, removed data elements may be employed as a key. In some embodiments, the removal and use of the unique data as an index key proceeds in a manner that is transparent to the data consuming user, process, or device.

Abstract: System and method for tracking a mobile device includes: receiving unique identifications for a mobile device; filtering out the unique identifications to obtain a true identifications for the mobile device; identifying cameras relevant to movement of the mobile device; receiving video streams; generating data structures for the video streams and tracking information of the mobile device, the data structure including time stamped videos, and viewpoints of the identified cameras; utilizing the data structures to retrieve, video and tracking information for the mobile device and the user, as the mobile moves in the site; and applying analytics to the retrieved video and tracking information to analyze behavior of the user and to predict what the user will do while on site.

Abstract: A relation detection model training solution. The relation detection model training solution mines freely available resources from the World Wide Web to train a relationship detection model for use during linguistic processing. The relation detection model training system searches the web for pairs of entities extracted from a knowledge graph that are connected by a specific relation. Performance is enhanced by clipping search snippets to extract patterns that connect the two entities in a dependency tree and refining the annotations of the relations according to other related entities in the knowledge graph. The relation detection model training solution scales to other domains and languages, pushing the burden from natural language semantic parsing to knowledge base population. The relation detection model training solution exhibits performance comparable to supervised solutions, which require design, collection, and manual labeling of natural language data.

Abstract: A device may receive one or more images captured by an image capture system. The one or more images may depict one or more objects. The device may process the one or more images using one or more image processing techniques. The device may identify the one or more objects based on processing the one or more images. The device may identify a context of the one or more images based on the one or more objects depicted in the one or more images. The device may determine whether the one or more objects contribute to a value of one or more metrics associated with the context. The device may perform an action based on the value of the one or more metrics.

Abstract: When blending individual vehicle-mounted camera images into a coherent composite image of the vehicle surroundings, objects in the respective images are analyzed to determine the presence of a pedestrian. When the individual images are blended to form the composite image, multiple pedestrian shadow views of the pedestrian are replaced with background pixels to erase the pedestrian, and a semi-symbolic graphical representation of the pedestrian is superimposed at the location of the actual pedestrian in the composite image. The graphical representation is proportional to the pedestrian in the composite image and includes features unique to the pedestrian it represents. The composite image with the graphical representation is then presented to a viewer.

Abstract: A method for storing XML documents a hybrid navigation/streaming format is provided to allow efficient storage and processing of queries on the XML data that provides the benefits of both navigation and streaming and ameliorates the disadvantages of each. Each XML document to be stored is independently analyzed to determine a combination of navigable and streamable storage format that optimizes the processing of the data for anticipated access patterns.

Abstract: A method includes receiving, at a processor of a computing device, a first search query during editing of a design, the first search query including at least a first search term. The method also includes identifying at least one style attribute associated with the design and searching at least one database based on the first search term, the at least one style attribute, or a combination thereof, to generate search results. The method further includes outputting the search results. A first search result of the search results is ranked higher than a second search result of the search results based on a determination that the first search result more closely matches the at least one style attribute associated with the design than the second search result.