Abstract: Systems and methods for training models to predict dense correspondences across images such as human images. A model may be trained using synthetic training data created from one or more 3D computer models of a subject. In addition, one or more geodesic distances derived from the surfaces of one or more of the 3D models may be used to generate one or more loss values, which may in turn be used in modifying the model's parameters during training.

Abstract: A method includes obtaining at least one image and a ground truth map associated with the at least one image. The method also includes generating multiple optical maps using multiple algorithms and the at least one image. The method further includes, for each algorithm, identifying at least one score for the algorithm using one or more of the optical maps generated using the algorithm and the ground truth map. The ground truth map identifies one or more boundaries associated with one or more foreground objects in the at least one image. The scores identify how well the optical maps generated using the algorithms separate the one or more foreground objects from a background in the at least one image.

Abstract: A method of tracking a container closure used to seal liquid in a container is provided, comprising applying a trackable indicia to a container closure, wherein the trackable indicia is unique to the container closure; recording an associated data set comprising a plurality of parameters related to the liquid in the container; correlating the associated data set with the trackable indicia; applying the container closure to the container to seal the liquid within the container; maintaining a database of the trackable indicia corresponding to the associated data set; and scanning the trackable indicia to retrieve the associated data set from the database. The method of the invention also includes utilization of a container closure having inherent natural imperfections and/or patterns on the top and/or sides of the closure and obtaining scanned images and data thereof which are analyzed and utilized for closure and container identification and likewise serve as scannable trackable indicia.

Abstract: Embodiments are disclosed for ground plane estimation (GPE) using a LiDAR semantic network. In an embodiment, a method comprises: obtaining a point cloud from a depth sensor of a vehicle operating in an environment; encoding the point cloud; estimating, using a deep learning network with the encoded point cloud as input, a ground plane in the environment; planning a path through the environment based on a drivable area of the estimated ground plane; and operating the vehicle, the vehicle along the path. The deep learning network includes a two-dimensional (2D) convolutional backbone, a detection head for detecting objects and a GPE head for estimating the ground plane. In an embodiment, point pillars are used to encode the point cloud.

Abstract: A system and a method for detecting a pile of material by an autonomous machine. A 3D point cloud method includes obtaining a 3D point cloud indicative of an environment having a material pile, performing a ground surface estimation on the point cloud to identify non-ground points, grouping the non-ground points into clusters based on a proximity metric, creating a normalized height histogram for each of the clusters, comparing the normalized height histogram of each cluster to a generalized pile histogram, and identifying a cluster as a pile based on the similarity between the normalized height histogram and the generalized pile histogram. A 2D image method includes obtaining a 2D image from an imaging device, calibrating the imaging device with respect to a coordinate frame of the machine, and autonomously detecting an image of a material pile in the two-dimensional image using a trained deep-learning neural network.

Abstract: A method for assisting operation of a vehicle traveling on a roadway includes acquiring visual images around the vehicle with at least one visual camera having a field of view and acquiring thermal images around the vehicle with at least one thermal camera having the field of view. The thermal images are superimposed over the visual images to produce composite images. An object is detected in the composite images. A vehicle assist system adjusts at least one of a direction of travel and speed of the vehicle in response to detecting the object.

Abstract: System for performing fully automatic brain tumor and tumor-aware cortex reconstructions upon receiving multi-modal MRI data (T1, T1c, T2, T2-Flair). The system outputs imaging which delineates distinctions between tumors (including tumor edema, and tumor active core), from white matter and gray matter surfaces. In cases where existing MRI model data is insufficient then the model is trained on-the-fly for tumor segmentation and classification. A tumor-aware cortex segmentation that is adaptive to the presence of the tumor is performed using labels, from which the system reconstructs and visualizes both tumor and cortical surfaces for diagnostic and surgical guidance. The technology has been validated using a publicly-available challenge dataset.

Abstract: A method for generating a trained neural network model for scanning correction corresponding to one or more imaging parameters is provided. The trained neural network model may be trained using training data. The training data may include at least one first set of training data. The first set of training data may be generated according to a process for generating the first set of training data. The process may include obtaining a first image and a second image corresponding to the one or more imaging parameters. The second image may include less scattering noises than the first image. The process may further include determine the first set of training data based on the first image and the second image.

Abstract: An image processing apparatus converts a multivalued image to a binary image, determines a character region and a non-character region in the binary image, extracts areas in units of a character from the character region, determines whether each of the areas is a character region or a non-character region, and extracts, for each of the areas determined to be the character region, a first representative color from the multivalued image included in the area, and generates a binary image corresponding to the first representative color. The apparatus decides a second representative color closest to a color of a target pixel by comparing the first representative color with the color of the target pixel, and changes the binary image of the area corresponding to the first representative color to a binary image of the decided second representative color.

Abstract: Systems and methods for identifying growth of a cerebral aneurysm. The method includes forming a first virtual skeleton model from a first image segmentation and forming a second virtual skeleton model from a second image segmentation. The virtual skeleton models including a plurality of edges with each edge having a plurality of skeleton points. Each skeleton point is associated with a subset of a plurality of blood vessel surface points. The method includes identifying one or more second terminal points within the second virtual skeleton model and overlapping the first virtual skeleton model and the second virtual skeleton model by orienting the one or more first terminal points with the one or more second terminal points.

Abstract: An information processing apparatus includes a character recognition unit that performs a character recognition process to recognize a character included in an image, and a partitioning unit that partitions a specific region if a character string recognized through the character recognition process performed by the character recognition unit on a specific region included in the image includes a character or a character string, each of which satisfies a predetermined condition.

Abstract: A banking system for associating an authorized payee with a customer account is described. The system includes a network interface, a memory, and a processor. The processor is configured to receive image data relating to a bill issued by a payee from a customer device via the network interface. The processor is further configured to perform optical character recognition of the image data. The processor is configured to extract a plurality of blocks of text from the image data. The processor is further configured to compare a first block of text of the image data to a database of known payees, wherein the first block of text relates to a name of the payee. The processor is configured to identify the payee from the database of known payees. The processor is further configured to associate the payee with the customer account.

Abstract: Computer systems and computer-implemented methods train a machine-learning regression system. The method comprises the step of generating, with a machine-learning generator, output patterns; distorting the output patterns of the generator by a scale factor to generate distorted output patterns; and training the machine-learning regression system to predict the scaling factor, where the regression system receives the distorted output patterns as input and learns and the scaling factor is a target value for the regression system. The method may further comprise, after training the machine-learning regression system, training a second machine-learning generator by back propagating partial derivatives of an error cost function from the regression system to the second machine-learning generator and training the second machine-learning generator using stochastic gradient descent.

Abstract: Embodiments are disclosed for ground plane estimation (GPE) using a LiDAR semantic network. In an embodiment, a method comprises: obtaining a point cloud from a depth sensor of a vehicle operating in an environment; encoding the point cloud; estimating, using a deep learning network with the encoded point cloud as input, a ground plane in the environment; planning a path through the environment based on a drivable area of the estimated ground plane; and operating the vehicle, the vehicle along the path. The deep learning network includes a two-dimensional (2D) convolutional backbone, a detection head for detecting objects and a GPE head for estimating the ground plane. In an embodiment, point pillars are used to encode the point cloud.

Abstract: Systems and methods for unmixing of multichannel image data in the presence of locally varying image characteristics. Feature images created from a multi-channel input image form a feature vector for each pixel in the input image. Feature vectors are classified based on the local image characteristics, and areas are formed in the input image that share local image characteristics. Each area is unmixed separately using reference vectors that were obtained from regions in example images that have the same image characteristics. The unmixing results are combined to form a final unmixing result image created from the individually unmixed image areas.

Abstract: Techniques for document segmentation. In an example, a document processing application segments an electronic document image into strips. A first strip overlaps a second strip. The application generates a first mask indicating one or more elements and element types in the first strip by applying a predictive model network to image content in the first strip and a prior mask generated from image content of the first strip. The application generates a second mask indicating one or more elements and element types in the second strip by applying the predictive model network to image content in the second strip and the first mask. The application computes, from a combined mask derived from the first mask and the second mask, an output electronic document that identifies elements in the electronic document and the respective element types.

Abstract: A method for dividing an image obtains the image by captured and a depth map of the image. The method creates a histogram of the depth map including horizontal and vertical axes. A clustering algorithm is applied to the data in the histogram to determine upon two data clusters and the cluster centers of the two clusters. The method determines that the abscissa value of the smallest ordinate value between two cluster centers in the histogram is a segmentation threshold. If the obtained segmentation threshold meets a preset condition, the image is segmented into a foreground region and a background region. An electronic device for applying the method is also provided.

Abstract: The present application relates to a method performed by an electronic device for positioning a table in a PDF document. The method comprises the following steps: receiving the PDF document containing the table; extracting character information and line information from vector stream information of the PDF document; and positioning a table area in the PDF document according to the extracted character information and line information. The method and the device in the present application perform table area positioning based on all lines and text blocks in a page. Compared with the prior art, the method and the device have the advantages that the accuracy of the table area positioning can be greatly improved, and a foundation is provided for accurate analysis of table information.

Abstract: An alignment method for manufacturing a mask integration framework is disclosed. The alignment method includes establishing an absolute coordinate system by taking a center of a metal framework as an origin of coordinates, the center of the metal framework coinciding with a center of an array substrate serving as a reference, controlling the array substrate to move, such that an offset of coordinates of a pixel point under the absolute coordinate system with respect to a predetermined theoretical value is smaller than or equal to a predetermined error value, and transmitting the coordinates of the pixel point under the absolute coordinate system, after the array substrate moves, to a tension device. An alignment system for manufacturing mask integration framework is also disclosed.

Abstract: A method for generating a trained neural network model for scanning correction corresponding to one or more imaging parameters is provided. The trained neural network model may be trained using training data. The training data may include at least one first set of training data. The first set of training data may be generated according to a process for generating the first set of training data. The process may include obtaining a first image and a second image corresponding to the one or more imaging parameters. The second image may include less scattering noises than the first image. The process may further include determine the first set of training data based on the first image and the second image.

Abstract: A supplemental device for attachment to an injection device, the supplemental device comprising: an imaging arrangement configured to capture an image of one or more numbers present on a moveable component of the injection device; and a processor arrangement configured to run a first algorithm to: receive data representing an image; determine the position of at least one number in the image by determining the boundaries of each visible number; select one of the visible numbers to be identified; perform a pattern matching process to identify the value of the selected number; and determine a dose dialled into the injection device using the determined position of the selected one of the at least one visible number in the image.

Abstract: Font replacement based on visual similarity is described. In one or more embodiments, a font descriptor includes multiple font features derived from a visual appearance of a font by a font visual similarity model. The font visual similarity model can be trained using a machine learning system that recognizes similarity between visual appearances of two different fonts. A source computing device embeds a font descriptor in a document, which is transmitted to a destination computing device. The destination compares the embedded font descriptor to font descriptors corresponding to local fonts. Based on distances between the embedded and the local font descriptors, at least one matching font descriptor is determined. The local font corresponding to the matching font descriptor is deemed similar to the original font. The destination computing device controls presentations of the document using the similar local font. Computation of font descriptors can be outsourced to a remote location.

Abstract: Apparatus, system and method for detecting defects in an adhesion area that includes an adhesive mixed with a fluorescent material. One or more illumination devices may illuminate the fluorescent material in the adhesion area with a light of a predetermined wavelength. A camera may be configured to capture an image of the illuminated adhesion area. A processing device, communicatively coupled to the camera, may be configured to process the captured image by applying one or more boundary areas to the captured image and determining an image characteristic within each of the boundary areas, wherein the image characteristic is used by the processing device to determine the presence of a defect in the adhesive, such as an excess of adhesive or an insufficient application of adhesive.

Abstract: A region-of-interest detection apparatus for improving detection accuracy of a region of interest in a case where a region of interest overlaps a background region, or in a case where feature amounts of a region of interest and a background region are similar to each other calculates feature amounts of regions where partial regions and a background region set in an input image overlap each other, and based on the calculated feature amounts and a feature amount of each position in the input image, calculates a foreground level of the position in the input image. Then, the region-of-interest detection apparatus detects a region of interest from the input image based on the calculated foreground level and a saliency of the position.

Abstract: Embodiments include a system and method for activity monitoring using video data from multiple dissimilar sources. The video data is processed to remove any dependency of the system on types of video input data. The video data is processed to yield useful human readable information regarding events in real time, such as how many people move through a line in a period of time.

Abstract: The present disclosure discloses method and text detection system for detecting text in digital engineering drawings. The text detection system receives digital engineering drawing, where digital image is converted to binary image. One or more lines inclined at one or more angles are detected based on pre-defined parameters, where one or more lines are removed from binary image. The text detection system determines image contours in binary image based on connected pixels with similar characteristics and identify one or more regions in binary image, to be non-text regions based on height, width, and pixel density, of corresponding image contours and predefined minimum height, minimum width, maximum height, maximum width, and predefined pixel density, where non-text regions are removed from binary image. The text detection system determines text regions in binary image based on removal of non-text regions and processing text regions to detect text in binary image of engineering drawing.

Abstract: Computer program products, methods, systems, apparatus, and computing entities provide a unique single-shot text detector that generates word-level text bounding boxes in an image by at least identifying text regions in the image via an automatically learned attention map and by conducting pixel-wise review of text; aggregating multi-scale inception features; generating, based at least in part on the multi-scale inception features, a set of aggregated inception features; and generating, using at least the set of aggregated inception features, the word-level text bounding boxes in the image.

Abstract: A method includes receiving, at a media device, overlay data associated with media content. The overlay data includes information indicating a time and a position at which a particular object is represented within the media content, and the overlay data includes information associating the particular object with supplemental information. The method further includes, during playback of the media content, receiving user input at the media device, and, responsive to the user input, generating an overlay display based on the overlay data, where the overlay display causes the particular object to be highlighted in a display of the media content.

Abstract: An image processing apparatus has a color image, the image data being constituted by multiple pixels, each of the multiple pixels having a gradation value, and a controller, which is configured to generate a histogram of index values corresponding to brightness values of the multiple pixels constituting the image data, set an original threshold value based on the histogram which is referred to for binarization, detect a mound-shaped part, in the histogram, satisfying a particular condition, set an adjusting direction in which the original threshold value is to be adjusted, set the index value at a base on a particular direction side of a particular mound-shaped part which is one of mound-shaped parts existing on the adjusting direction side with respect to the original threshold value in the histogram as an adjusted threshold value, and apply a binarizing process to the image data using the adjusted threshold value.

Abstract: A method is described that includes translating higher level program code including higher level instructions having an instruction format that identifies pixels to be accessed from a memory with first and second coordinates from an orthogonal coordinate system into lower level instructions that target a hardware architecture having an array of execution lanes and a shift register array structure that is able to shift data along two different axis. The translating includes replacing the higher level instructions having the instruction format with lower level shift instructions that shift data within the shift register array structure.

Abstract: A method for processing an electronic document with characters includes adjusting the characters to identify lines and words; generating a cluster encompassing all of the lines and the words; setting the cluster as a target; determining whether the target can be divided; in response to determining that the target can be divided, dividing the target into a first plurality of sub-clusters; identifying blocks of text based on the first sub-clusters; and generating a new electronic document with paragraphs and sections based on the blocks of text.

Abstract: Provided is a segment-block-based handwritten signature authentication system and a method thereof, and more particularly, to a handwritten signature authentication system and a method thereof that enrolls a handwritten signature including handwritten signature characteristics information based on segment blocks disjointed by a user, acquires segment-block-based handwritten signature characteristics information from the handwritten signature upon request for handwritten signature authentication, and performs handwritten signature authentication by comparing the pre-enrolled handwritten signature characteristics information based on segments and the acquired handwritten signature characteristics information.

Abstract: A method of encoding display output of a computer using a block-based video encoder, such as H.264 includes receiving display update information, the display update information specifying at least one changed display area. Pixel data associated with the at least one changed area is also received. For each image block of an output frame to be encoded, the encoder determines whether the image block intersects one or more of the changed areas. A block skip indication is encoded for each image block not intersecting at least one changed area, and pixel data is encoded for intersecting blocks. An encoded output frame is then generated including the block skip indications and encoded pixel data.

Abstract: A system and method for protecting copyright in content distributed online, in combination with specified business rules. A portion of content presented for upload on a network is analyzed to detect an image associated with a content owner; the image is compared with reference images to identify the content owner; and business rules are applied to control unauthorized uploading of the content. The identifier may be a logo included in the content as a digital graphic, or a non-visual marker. Analysis is advantageously performed on a sample of video frames or a segment of preselected length. If the content is found to be copyrighted, and the attempted upload is unauthorized, uploading may or may not be permitted, and the user may or may not be charged a fee for subsequent access to the content.

Abstract: The disclosure relates to a method and device for processing a page. The method includes calculating an average brightness value of pixels in a page on a display; determining whether the average brightness value exceeds a preset brightness threshold; selecting target pixels from the page when the average brightness value exceeds the preset brightness threshold, wherein the target pixels include at least one of first target pixels and second target pixels, the first target pixels including sub-pixels, any two of which have a brightness difference that is less than a preset difference value, and the second target pixels including background pixels and character pixels; and performing a color reversion operation on the target pixels.

Abstract: Transient connected regions are tracked in a video sequence, marking them as foreground layer or background layer when the transient regions become stable. A stack of background/foreground Gaussian mixer models is maintained for each pixel. The decision to mark a stable region as background layer or a new or existing foreground layer is done by matching the region with each model in the model stack. If the new region matches an existing model, then the layers above the matched layers are purged, or else the new region is pushed as a new foreground layer with a new model.

Abstract: The method of estimating a quality index of a gray-scale encoded three-dimensional image of a composite material part includes an obtaining step for obtaining a histogram from the image, the histogram representing for each class of a plurality of classes each including at least one gray-scale level, the number of voxels of the image having a gray-scale level belonging to the class; an extraction step for extracting a predetermined number of Gaussian functions present in the histogram; and an estimation step for estimating at least one quality index of the image from parameters characterizing the Gaussian functions extracted from the histogram.

Abstract: Described herein are systems and methods for formatting text for generating a print document based on a web document. For example, in some examples, a subset of characters of a first element of a web document is wrapped in a second element. Further, the onscreen position of the subset is determined based on the wrapping. In addition, the position on a print document for the subset is determined based on the determined onscreen position.

Abstract: Embodiments of a method and system described herein enable capture of video data streams from multiple, different video data source devices and the processing of the video data streams. The video data streams are merged such that various data protocols can all be processed with the same worker processors on different types of operating systems, which are typically distributed.

Abstract: A processing apparatus includes a photographed image obtainer configured to obtain a plurality of photographed images of an object, the images being photographed by sequentially irradiating light onto the object from a plurality of different light source positions, a first shadow map generator configured to generate a first shadow map by determining a shadow area in the plurality of photographed images, a second shadow map generator configured to generate a second shadow map based on the first shadow map, and a normal information obtainer configured to obtain normal information of the object calculated based on the second shadow map and the plurality of photographed images.

Abstract: Methods, devices, and non-transitory computer readable storage media for extracting text from documents are disclosed. The method includes performing layout analysis on the document to identify a plurality of regions within a plurality of pages in the document. The method further includes identifying a table region from within the plurality of regions based on homogeneity between a plurality of textual lines in a page from the plurality of pages. The method includes identifying at least two rows and at least two columns within the table region. The method further includes identifying a plurality of cells within the table region based on the at least two rows and the at least two columns. The method includes extracting text from each of the plurality of cells.

Abstract: Systems, apparatuses and methods may provide away to render augmented reality and virtual reality (VR/AR) environment information. More particularly, systems, apparatuses and methods may provide a way to selectively suppress and enhance VR/AR renderings of n-dimensional environments. The systems, apparatuses and methods may deepen a user's VR/AR experience by focusing on particular feedback information, while suppressing other feedback information from the environment.

Abstract: Apparatus, system and method for detecting defects in an adhesion area that includes an adhesive mixed with a fluorescent material. One or more illumination devices may illuminate the fluorescent material in the adhesion area with a light of a predetermined wavelength. A camera may be configured to capture an image of the illuminated adhesion area. A processing device, communicatively coupled to the camera, may be configured to process the captured image by applying one or more boundary areas to the captured image and determining an image characteristic within each of the boundary areas, wherein the image characteristic is used by the processing device to determine the presence of a defect in the adhesive, such as an excess of adhesive or an insufficient application of adhesive.

Abstract: Methods and systems for selecting listing based on a click predictions score are disclosed. The server system includes one or more processors, memory, and one or more programs stored in the memory. The server system receives a search query from a client system. In response to receiving the search query, the server system identities a plurality of item listings relevant to the received search query. For each identified item listing, the server system generates a click prediction score, wherein a click predication score represents the estimated likelihood that a user will click on the listing. The server system ranks the plurality of item listings based on the generated click prediction score and selects a predetermined number of the item listings for transmission to the client system based on their ranking and associated click prediction score. The server system transmits one or more item listings to the client system for display.

Abstract: A method for determining an orientation of a screen of an electronic apparatus and an image receiving apparatus are provided. The method for determining the screen orientation includes the following steps of: receiving a screen shared image transmitted by the electronic apparatus, the screen shared image is a color image; performing line sampling on pixels of the screen shared image; and determining whether the screen is in the landscape orientation or in the portrait orientation according to color values of the sampled pixels of the screen shared image. The present disclosure can determine the screen orientation according to color values of the sampled pixels with small amount of operation data and high accuracy.

Abstract: A method, non-transitory computer readable medium, and apparatus for providing a customized handwriting zone are disclosed. For example, the method receives a handwriting sample for a user, measures dimensions of the handwriting sample for the user, generates one or more handwriting zones that are sized in accordance with the dimensions of the handwriting sample of the user, receives a request to generate a form for the user and provides the form that includes the one or more handwriting zones that are customized in accordance with the dimensions of the handwriting sample of the user.

Abstract: The claimed invention relates to the method of data entering and displaying on devices equipped with touch screens using the Graphic User Interface (GUI) comprising a virtual keyboard providing interaction of a user with device. The claimed effect is reached due to the method of entering and displaying of alphanumeric and other graphical symbols on the electronic device touch screen comprising the following steps: generate on a device touch screen a virtual keyboard consisting of data input/output virtual keys, and each virtual key comprises the input symbol sample and the field of displaying of the entered data; determine user interaction with the said virtual keyboard; and arrange and display on the device display in the virtual key field of displaying of the entered data symbols entered by the user using the said virtual keyboard.

Abstract: The invention relates to the optical detection of symbols (4) displayed on a display (2), wherein image data which contains a representation of the display (2) is obtained, comprising the steps: detection (29, 30) of a display area of the display (2) in the obtained image data; preparation (32) of the image data representing the detected display area; performing an optical symbol detection (34, 35) by using the prepared image data; and storing the symbols detected during the optical symbol detection (34, 35) in a memory (13).

Abstract: A method includes: displaying a digital image on a first portion of a display of a mobile device; receiving user feedback via the display of the mobile device; analyzing the user feedback to determine a meaning of the user feedback; based on the determined meaning of the user feedback, analyzing a portion of the digital image corresponding to either the point of interest or the region of interest to detect one or more connected components depicted within the portion of the digital image; classifying each detected connected component depicted within the portion of the digital image; estimating an identity of each detected connected component based on the classification of the detected connected component; and one or more of: displaying the identity of each detected connected component on a second portion of the display of the mobile device; and providing the identity of each detected connected component to a workflow.

Abstract: A supplemental device (2) for attachment to a pen-type drug injection device, the supplemental device comprising an optical sensor array or camera (25) configured to capture an image of one or more numbers representing dose values on a dose dial sleeve of the injection device, and a processor (24) and memory (240,241) arrangement configured to run an algorithm to receive image data from the sensor/camera and extract the dialed dose value from said image via optical character recognition (OCR). The algorithm is able to determine the set dose value which is at the centre of the image (902), also if the dose scale only comprises integer values in steps of two (see FIG. 9).