Abstract: Systems and methods are disclosed for training and using neural networks for computing depth maps. One method for training the neural network includes providing an image input to the neural network. The image input may include a camera image of a training scene. The method may also include providing a depth input to the neural network. The depth input may be based on a high-density depth map of the training scene and a sampling mask. The method may further include generating, using the neural network, a computed depth map of the training scene based on the image input and the depth input. The method may further include modifying the neural network based on an error between the computed depth map and the high-density depth map.

Abstract: The purpose of the present invention is to provide a pattern matching device and computer program that carry out highly accurate positioning even if edge positions and numbers change. The present invention proposes a computer program and a pattern matching device wherein a plurality of edges included in first pattern data to be matched and a plurality of edges included in second pattern data to be matched with the first pattern data are associated, a plurality of different association combinations are prepared, the plurality of association combinations are evaluated using index values for the plurality of edges, and matching processing is carried out using the association combinations selected through the evaluation.

Abstract: A method and apparatus for image processing. A de-warping circuit is configured to retrieve, in an ordered sequence, a plurality of de-warping matrices for remapping pixels of a first image from a distorted image space to a corrected image space, where the first image depicts a scene in the distorted image space. The de-warping circuit reads, into a memory buffer, blocks of pixels of the first image based at least in part on the order of the de-warping matrices and generates a plurality of image tiles based on the blocks of pixels of the first image. Each of the image tiles is interpolated from one or more of the retrieved blocks using a respective one of the de-warping matrices. The de-warping circuit writes the plurality of image tiles to an external memory to produce a second image depicting the scene in the corrected image space.

Abstract: Embodiments herein provide a method for stereo-visual localization of an object by a stereo-visual localization apparatus. The method includes generating, by a stereo-visual localization apparatus, a stereo-visual interface displaying the first stereo image of the object and the first stereo image of the subject in a first portion and the second stereo image of the object and the second stereo image of the subject in a second portion. Further, the method includes detecting, by the stereo-visual localization apparatus, a movement of the subject to align the subject in the field of view with the object. Furthermore, the method includes visually aligning, by the stereo-visual localization apparatus, the subject with the object based on the movement by simultaneously changing apparent position of the first and the second stereo images of the subject in each of the first portion and the second portion in the stereo-visual interface.

Abstract: A method for detecting a user input based on a gesture in which image data of at least two individual images are acquired and recoding times are allocated to the individual images. Each of the acquired individual images is segmented, an individual image object is identified in each of the individual images and a reference point is determined based on the individual image object. A trajectory is determined based on the reference points in the individual images and a gesture is determined based on the trajectory. An output signal is generated and output based on the gesture determined. A device for detecting a user input based on a gesture having an acquisition unit for acquiring image data, a segmentation unit for performing segmentation, a trajectory computing unit for determining the trajectory, an allocation unit for determining a gesture and an output unit.

Abstract: Systems and methods for tracking the location of a non-destructive inspection (NDI) scanner using images of a target object acquired by the NDI scanner. The system includes a frame, an NDI scanner supported by the frame, a system configured to enable motorized movement of the frame, and a computer system communicatively coupled to receive sensor data from the NDI scanner and track the location of the NDI scanner. The NDI scanner includes a two-dimensional (2-D) array of sensors. Subsurface depth sensor data is repeatedly (recurrently, continually) acquired by and output from the 2-D sensor array while at different locations on a surface of the target object. The resulting 2-D scan image sequence is fed into an image processing and feature point comparison module that is configured to track the location of the scanner relative to the target object using virtual features visible in the acquired scan images.

Abstract: Variation in received documents types and templates used for each document type poses challenge in developing a generic background noise removal approach for automatic text information extraction technique. Embodiments herein provide a method and a system for document classification and text information extraction. Time efficient and accurate text detection engine-based Region of Interest (ROI) technique is provided to accurately identify text region followed by a multi-layered neural network based architecture for enhanced classification accuracy to identify the type of document. A multistage image pre-processing approach is provided for efficient, effective, and accurate background noise removal from the classified document, which includes unsupervised clustering, identification, segmentation, masking, contour approximation, selective subtraction, and dynamic thresholding.

Abstract: An image processing apparatus includes: a model pattern storage unit that stores a model pattern composed of a plurality of model feature points; an image data acquisition unit that acquires a plurality of images obtained through capturing an object to be detected; an object detection unit that detects the object to be detected from the images using the model pattern; a model pattern transformation unit that transforms a position and posture such that the model pattern is superimposed on an image of the object to be detected; a corresponding point acquisition unit that acquires a corresponding point on image data corresponding to each of the model feature points; a corresponding point set selection unit that selects a set of corresponding points on the plurality of images; and a three-dimensional position calculation unit that calculates a three-dimensional position of the image of the object to be detected.

Abstract: The present invention discloses fiducial marker systems or tag systems and methods to detect and decode a tag. In one aspect, a tag comprises four corners. Two upper corners are interconnected to form a detection area. Two lower corners are interconnected to form another detection area. The detection areas are interconnected by a path. The path divides the space between the detection areas into two coding areas. In another aspect, a tag comprises four corners. The four corners are interconnected by multiple paths. The multiple paths divide the space defined by the four corners into multiple coding areas.

Abstract: Systems and methods for generating synthesized medical images for training a machine learning based network. An input medical image in a first modality is received comprising a nodule region for each of one or more nodules, a remaining region and an annotation for each of the nodules. A synthesized medical image in a second modality is generated from the input medical image comprising the annotation for each of the nodules. A synthesized nodule image of each of the nodule regions and synthesized remaining image of the remaining region are generated in the second modality. It is determined whether a particular nodule is visible in the synthesized medical image based on the synthesized nodule image for the particular nodule and the synthesized remaining image. If at least one nodule is not visible in the synthesized medical image, the annotation for the not visible nodule is removed from the synthesized nodule image.

Abstract: An image synthesis unit of an X-ray imaging apparatus is configured to correct a synthesis target image or a transparent image based on movement information of a feature point and movement information of a pixel and generate a synthesized image by synthesizing a corrected synthesis target image and a transparent image or synthesizing a synthesis target image and a corrected transparent image.

Abstract: Systems and methods of script identification in audio data obtained from audio data. The audio data is segmented into a plurality of utterances. A script model representative of a script text is obtained. The plurality of utterances are decoded with the script model. A determination is made if the script text occurred in the audio data.

Abstract: Implementations of the subject matter described herein provide a solution for extracting spatial-temporal feature representation. In this solution, an input comprising a plurality of images is received at a first layer of a learning network. First features that characterize spatial presentation of the images are extracted from the input in a spatial dimension using a first unit of the first layer. Based on a type of a connection between the first unit and a second unit of the first layer, second features at least characterizing temporal changes across the images are extracted from the first features and/or the input in a temporal dimension using the second unit. A spatial-temporal feature representation of the images is generated partially based on the second features. Through this solution, it is possible to reduce learning network sizes, improve training and use efficiency of learning networks, and obtain accurate spatial-temporal feature representations.

Abstract: A method for measuring a seam on aircraft skin based on a large-scale point cloud is disclosed. A point cloud density of each point in an aircraft skin point cloud is calculated. Seam and non-seam point clouds are divided according to a discrepancy of the calculated point cloud density. A point is selected from the point cloud of the seam area, and a section at the point is extracted. A certain range of the seam and non-seam point clouds is projected to the section and a projected point cloud is acquired. A calculation model of flush and gap is constructed, and the flush and the gap of the aircraft skin seam at the measuring point is calculated according to the projected point cloud and the calculation model.

Abstract: This disclosure is directed to a system and method that automatically detects repeated content within multiple media items. Content providers often include content, such as an introduction, near the beginning of a media item. In some circumstances, such as in the case of a series of television episodes, the content providers use the same content in each episode of the series. By dividing the media items into portions and analyzing the portions, the systems and methods described can automatically detect the repeated content. Using the detection of the repeated content, a user interface can then allow a user to bypass the repeated content during playback.

Abstract: Systems and methods of script identification in audio data obtained from audio data. The audio data is segmented into a plurality of utterances. A script model representative of a script text is obtained. The plurality of utterances are decoded with the script model. A determination is made if the script text occurred in the audio data.

Abstract: Disclosed are methods, apparatuses and systems for CNN network adaption and object online tracking. The CNN network adaption method comprises: transforming a first feature map into a plurality of sub-feature maps, wherein the first feature map is generated by the pre-trained CNN according to a frame of the target video; convolving each of the sub-feature maps with one of a plurality of adaptive convolution kernels, respectively, to output a plurality of second feature maps with improved adaptability; training, frame by frame, the adaptive convolution kernels.

Abstract: Input images are partitioned into non-overlapping segments perpendicular to a disparity dimension of the input images. Each segment includes a contiguous region of pixels spanning from a first edge to a second edge of the image, with the two edges parallel to the disparity dimension. In some aspects, contiguous input image segments are assigned in a “round robin” manner to a set of sub-images. Each pair of input images generates a corresponding pair of sub-image sets. Semi-global matching processes are then performed on pairs of corresponding sub-images generated from each input image. The SGM processes may be run in parallel, reducing an elapsed time to generate respective disparity sub-maps. The disparity sub-maps are then combined to provide a single disparity map of equivalent size to the original two input images.

Abstract: A method for automatically classifying emission tomographic images includes receiving original images and a plurality of class labels designating each original image as belonging to one of a plurality of possible classifications and utilizing a data generator to create generated images based on the original images. The data generator shuffles the original images. The number of generated images is greater than the number of original images. One or more geometric transformations are performed on the generated images. A binomial sub-sampling operation is applied to the transformed images to yield a plurality of sub-sampled images for each original image. A multi-layer convolutional neural network (CNN) is trained using the sub-sampled images and the class labels to classify input images as corresponding to one of the possible classifications. A plurality of weights corresponding to the trained CNN are identified and those weights are used to create a deployable version of the CNN.

Abstract: Provided are systems and methods to perform colour extraction from swatch images and to define new images using extracted colours. Source images may be classified using a deep learning net (e.g. a CNN) to indicate colour representation strength and drive colour extraction. A clustering classifier is trained to use feature vectors extracted by the net. Separately, pixel clustering is useful when extracting the colour. Cluster count can vary according to classification. In another manner, heuristics (with or without classification) are useful when extracting. Resultant clusters are evaluated against a set of (ordered) expected colours to determine a match. Instances of standardized swatch images may be defined from a template swatch image and respective extracted colours using image processing. The extracted colour may be presented in an augmented reality GUI such as a virtual try-on application and applied to a user image such as a selfie using image processing.