7126 results about "Image pair" patented technology

Depth maps are generated from a monoscopic source images and asymmetrically smoothed to a near-saturation level. Each depth map contains depth values focused on edges of local regions in the source image. Each edge is defined by a predetermined image parameter having an estimated value exceeding a predefined threshold. The depth values are based on the corresponding estimated values of the image parameter. The depth map is used to process the source image by a depth image based rendering algorithm to create at least one deviated image, which forms with the source image a set of monoscopic images. At least one stereoscopic image pair is selected from such a set for use in generating different viewpoints for multiview and stereoscopic purposes, including still and moving images.

A system and method for improving digital flash photographs. The present invention is a technique that significantly improves low-light imaging by giving the end-user all the advantages of flash photography without producing the jarring look. The invention uses an image pair—one taken with flash the other without—to remove noise from the ambient image, sharpen the ambient image using detail from the flash image, correct for color, and remove red-eye.

<heading lvl="0">Abstract of Disclosure</heading> This invention relates to an electrophoretic display or a liquid crystal display and novel processes for its manufacture. The electrophoretic display (EPD) of the present invention comprises microcups of well-defined shape, size and aspect ratio and the microcups are filled with charged pigment particles dispersed in an optically contrasting dielectric solvent. The liquid crystal display (LCD) of this invention comprises well-defined microcups filled with at least a liquid crystal composition having its ordinary refractive index matched to that of the isotropic cup material. A novel roll-to-roll process and apparatus of the invention permits the display manufacture to be carried out continuously by a synchronized photo-lithographic process. The synchronized roll-to-roll process and apparatus permits a pre-patterned photomask, formed as a continuous loop, to be rolled in a synchronized motion in close parallel alignment to a web which has been pre-coated with a radiation sensitive material, so as to maintain image alignment during exposure to a radiation source. The radiation sensitive material may be a radiation curable material, in which the exposed and cured portions form the microcup structure. In an additional process step, the radiation sensitive material may be a positively working photoresist which temporarily seals the microcups. Exposure of a selected subset of the microcups via the photomask image permits selective re-opening, filling and sealing of the microcup subset. Repetition with additional colors permits the continuous assembly of a multicolor EPD or LCD display.

A “Keyframe Stitcher” provides an efficient technique for building mosaic panoramic images by registering or aligning video frames to construct a mosaic panoramic representation. Matching of image pairs is performed by extracting feature points from every image frame and matching those points between image pairs. Further, the Keyframe Stitcher preserves accuracy of image stitching when matching image pairs by utilizing ordering information inherent in the video. The cost of searching for matches between image frames is reduced by identifying “keyframes” based on computed image-to-image overlap. Keyframes are then matched to all other keyframes, but intermediate image frames are only matched to temporally neighboring keyframes and neighboring intermediate frames to construct a “match structure.” Image orientations are then estimated from this match structure and used to construct the mosaic. Matches between image pairs may be compressed to reduce computational overhead by replacing groups of feature points with representative measurements.

The present invention is an imaging device which can be equipped with decode functionality and which can be configured to operate in at least one of four user selected modes. In a first, “message only” mode, the device stores into a designated memory location a decoded-out message corresponding to a decodable indicia. In a second, “image only” mode, the device stores into a designated frame storage memory location an image representation of a scene. In a third, “omage plu message” mode the device stores into a designated frame storage memory location an image representation comprising a representation of a decodable symbol and into the same or other memory location the decoded-out message decoded from the decodable indicia, or data corresponding to the same. In the fourth, “two-step message and image” mode, the device is controlled to capture an image a first time for decoding a decodable indicia and a second time for capturing an image that is associated with the decoded-out message corresponding to the decodable indicia.

A system and method are provided for autonomously navigating a vehicle. The method captures a sequence of image pairs using a stereo camera. A navigation application stores a vehicle pose (history of vehicle position). The application detects a plurality of matching feature points in a first matching image pair, and determines a plurality of corresponding object points in three-dimensional (3D) space from the first image pair. A plurality of feature points are tracked from the first image pair to a second image pair, and the plurality of corresponding object points in 3D space are determined from the second image pair. From this, a vehicle pose transformation is calculated using the object points from the first and second image pairs. The rotation angle and translation are determined from the vehicle pose transformation. If the rotation angle or translation exceed a minimum threshold, the stored vehicle pose is updated.

A method of estimating three-dimensional camera position information from a series of two-dimensional images that form a panorama employs common features in adjoining image pairs in the series to estimate a transform between the images in the pairs. The common features are subsequently employed to adjust an estimated rotational component of each transform by reducing error between coordinates corresponding to the common features in three-dimensional space in image pairs, on a pair-by-pair basis. A global optimization of the position estimation, used for long sequences of images such as 360 degree panoramas, refines the estimates of the rotational and focal length components of the transforms by concurrently reducing error between all 3D common feature coordinates for all adjoining pairs.

Systems, methods and articles of manufacture are disclosed for stereoscopically editing video content. In one embodiment, image pairs of a sequence may be stereoscopically modified by altering at least one image of the image pair. The at least one image may be altered using at least one mapping function. The at least one image may also be altered based on a saliency of the image pair. The at least one image may also be altered based on disparities between the image pair. Advantageously, stereoscopic properties of video content may be edited more conveniently and efficiently.

A method for rectifying misalignment in a stereoscopic display system (140) comprises: providing a pair of input images to an image processor (120); creating an image source displacement map for the pair of input images; obtaining a display displacement map (150); and applying the image source displacement map and the display displacement map to the pair of input images to create a rectified stereoscopic image pair.

Methods and Systems are provided for converting a two-dimensional image sequence into a three-dimensional image. In one embodiment, a method for converting a two-dimensional image sequence into a three-dimensional image includes determining camera motion parameters between consecutive images in a monoscopic sequence of reference 2D images, wherein the consecutive images comprise a current reference image and an adjacent image, determining a horizontal disparity map for a target image using the camera motion parameters, determining a disparity probability value for each disparity vector of the disparity map, and determining a target image as a weighted average of pixel values in the current reference image using the disparity probability values, such that the target image and current reference image comprise a stereoscopic image pair.

An adaptive fill-flash technique for digital photography that employs flash and no-flash image pairs. A flash image, captured using a flash of known characteristics and ambient light, and an ambient light image, which is taken with only ambient light, of the same image subject matter are acquired. A flash only image, computed by subtracting the known ambient light image from the flash and ambient light image, is computed. The flash only image and the ambient light image are then each white balanced to the flash light and the ambient light, respectively, to form two white-balanced images. These two white-balanced images are then added to form a white balanced image, I, in which the flash and ambient light color agree. It is also possible to reduce the hot spots in image I, and adjust flash strength.

A pose-aware feature learning system includes an object tracker which tracks an object on a subject in a plurality of video frames, a pose estimator which estimates a pose of the subject in a track of the plurality of video frames, an image pair generator which extracts a plurality of image pairs from the track of the plurality of video frames, and labels the plurality of image pairs with the estimated pose and as depicting the same or different object, and a neural network trainer which trains a neural network based on the labeled plurality of image pairs, to predict whether an image pair depicts the same or different object and a pose difference for the image pair.

In a method, device and storage medium encoded with programming instructions for automatic image registration of image data of a current medical image MR study and at least one reference study, corresponding image pairs of the current study and the reference study are formed automatically with an association machine without needing the analyze the respective image data or pixel data. The pair determination takes place exclusively on the basis of the DICOM header data. A synchronized image processing and/or presentation of the generated image pairs takes place at a monitor.

The invention discloses an infrared target instance segmentation method based on feature fusion and a dense connection network, and the method comprises the steps: collecting and constructing an infrared image data set required for instance segmentation, and obtaining an original known infrared tag image; Performing image enhancement preprocessing on the infrared image data set; Processing the preprocessed training set to obtain a classification result, a frame regression result and an instance segmentation mask result graph; Performing back propagation in the convolutional neural network by using a random gradient descent method according to the prediction loss function, and updating parameter values of the convolutional neural network; Selecting a fixed number of infrared image data training sets each time and sending the infrared image data training sets to the network for processing, and repeatedly carrying out iterative updating on the convolutional network parameters until the convolutional network training is completed by the maximum number of iterations; And processing the test set image data to obtain average precision and required time of instance segmentation and a finalinstance segmentation result graph.

The invention discloses a target tracking method and system of a convolution twin network based on multi-layer feature fusion. The method comprises the following steps of: according to the target position and size of the images, cutting out the target template images and the search area images of all images in the image sequence training set, and forming a training data set by image pairs composedof the target template images and the search area images; constructing a convolution twin network based on multi-layer feature fusion; training the convolution twin network based on the multi-layer feature fusion based on the training data set to obtain a trained convolution twin network based on the multi-layer feature fusion; using the trained convolution twin network based on multi-layer feature fusion for target tracking. In the process of tracking targets, the invention integrates scores of different layers, combines high-level semantic features and bottom-level detail features, better distinguishes the interference of similar or similar targets, and prevents the problems of target drift and target loss in the tracking process.

A method and apparatus for motion blur and ghosting prevention in imaging system is presented. A residue image is computed by performing spatial-temporal filter with a set of absolute image difference of image pairs from input images. A noise adaptive pixel threshold is computed for every pixel based on noise statistics of image sensor. The residue image and the noise adaptive pixel threshold are used to create a motion masking map. The motion masking map is used to represent motion and non-motion pixels in pixels merging. The pixels merging step is performed to generate an output image by considering the motion pixels where the motion pixels are performed separately. The resulting output image having no or less motion blur and ghosting artifacts can be obtained, even the input images having different degree of motion blur between each of the image, while the complexity is low. It is preferred that the current invention is applied in the Bayer raw domain. The benefit is reduced computation and memory because only 1 color component is processed for each pixel. Another benefit is higher signal fidelity because processing in the Bayer raw domain is unaffected by demosaicing artifacts, especially along edges. However, the current invention can also be applied in RGB domain.

A method reduces artifacts in an input image. A variance image is generated from the input image. The input image is partitioned into a plurality of blocks of pixels. A set of classifications is defined. The set of classifications includes smooth, texture, and edge. A particular classification is assigned to each block of pixels of the input image according to the variance image, to generate smooth blocks, texture blocks, and edge blocks. A fuzzy filter is applied to each pixel of only each edge block.

The invention discloses a target posture measuring method based on binocular vision under double mediums, mainly solving the problem that in an underwater spacecraft simulation experiment, the posture of a target spacecraft cannot be measured. The target posture measuring method specifically comprises the following steps of: collecting a target spacecraft image pair through a binocular camera; carrying out harris angle point detection on the collected left image to find out a projection characteristic point; utilizing an epipolar constraint rule and a pyramid rapid matching method to find out an image projection characteristic of the right image; calculating three-dimensional coordinates of corresponding space characteristic points according to the projection characteristics of the left and right images; establishing a refraction model and revising the three-dimensional coordinates of the space characteristic points according to the refraction model; and screening characteristic points in the same plane and accurately calculating the posture of the target spacecraft according to the three-dimensional coordinates of the characteristic points. When the target posture measuring method is used in an underwater spacecraft simulation experiment, the three-dimensional coordinates of the characteristic points on the target spacecraft are accurately calculated through establishment of the refraction model, and the posture of the spacecraft can be accurately measured and calculated.

Foreground object matting uses flash/no-flash images pairs to obtain a flash-only image. A trimap is obtained from the flash-only image. A joint Bayesian algorithm uses the flash-only image, the trimap and one of the image of the scene taken without the flash or the image of the scene taken with the flash to generate a high quality matte that can be used to extract the foreground from the background.

A 2D or 3D velocity field is reconstructed from a cross-correlation analysis of image pairs of a sample, without first reconstructing images of the sample spatial structure. The method can be implemented via computer tomographic x-ray particle image velocimetry, using multiple projection angles, with phase contrast images forming dynamic speckle patterns. Estimated cross-correlations may be generated via convolution of a measured autocorrelation function with a velocity probability density function, and the velocity coefficients iteratively optimised to minimise the error between the estimated cross-correlations and the measured cross-correlations. The method may be applied to measure blood flow, and the motion of tissue and organs such as heart and lungs.

The present invention provides a method for detecting human faces, a device for detecting human faces and a remote teller machine system. The method includes the following steps that: an image pair to be recognized is acquired, wherein the image pair to be recognized includes two images to be recognized which are acquired by two cameras according to a human face to be recognized respectively; the depth information of the human face to be recognized is obtained according to the images to be recognized; a light spot pattern formed by the human face to be recognized under the illumination of an infrared structure is obtained; the texture information of the human face to be recognized is obtained according to the light spot pattern; and whether the human face to be recognized is a living body is determined based on the depth information and the texture information. The method, the device and the remote teller machine system have the advantages of low cooperation requirements, high speed and high security.

A method of rectifying a stereoscopic image consisting of left and right captured images comprises determining left and right rectification transformations. According to one aspect of the invention, statistics of the parameters of the stereoscopic image capture device used to capture the left and right images are used in the determination of the left and/or right rectification transformation.

According to another aspect of the invention, the left and right rectification transformations are constrained to correspond to a transformation to a virtual alignment to a parallel camera set-up.

Once the left and right rectification transformations have been determined they are preferably used to rectify the left and right images to eliminate, or substantially eliminate, vertical disparity from the rectified image pair. The left and right rectified images may then be displayed on a stereoscopic display device for viewing by an observer, or they may alternatively be stored for later use.

The invention discloses a depth image estimating method of a binocular stereo video, which comprises the following steps of: (1) collecting a binocular stereo video image from a binocular stereo camera; (2) performing parallax estimation based on graph cut stereo matching on the binocular stereo video image; (3) performing uniformity check on a parallax image obtained by the graph cut stereo matching, and deleting the unreliable matching by self-adapting matching to reduce the wrong matching of the depth image; (4) converting the parallax image into a depth image according to the relation of the parallax and the depth; (5) performing correction optimization on the obtained depth image by a multilateral filter; and (6) outputting the depth image to finish the depth image estimation of the binocular stereo video. The method disclosed by the invention has the advantages of effectively eliminating the errors in the depth image estimation, and finally obtaining an accurate and dense depth image so as to satisfy the demand of rebuilding image quality based on a real scene.

The invention discloses a multi-visual-angle face comparison method which comprises the steps of an acquisition step of acquiring three-dimensional face data of a person and performing face gesture correction for determining the front gesture of the three-dimensional face and forming a three-dimensional face model; rotating the three-dimensional face model after gesture correction around an X axis and a Y axis according to a preset angle interval, and performing perspective projection for obtaining a corresponding multi-visual-angle two-dimensional face image under a plurality of angles; performing normalization processing on the multi-visual-angle two-dimensional face image obtained through projection, then extracting characteristics for establishing a multi-visual-angle two-dimensional face template database and storing the characteristics into a registration database; and a comparison step, inputting a two-dimensional face image, comparing the input two-dimensional face image with each face with a similar gesture angle in the registration database, thereby identifying or verifying the identity of the two-dimensional face image.

The invention discloses a real-time dense monocular simultaneous localization and mapping (SLAM) method based on an online learning depth prediction network. The method comprises: optimization of a luminosity error of a minimized high gradient point is carried out to obtain a camera attitude of a key frame and the depth of the high gradient point is predicted by using a trigonometric survey methodto obtain a semi-dense map of a current frame; an online training image pair is selected, on-line training and updating of a CNN network model are carried out by using a block-by-block stochastic gradient descent method, and depth prediction is carried out on the current frame of picture by using the trained CNN network model to obtain a dense map; depth scale regression is carried out based on the semi-dense map of the current frame and the predicted dense map to obtain an absolute scale factor of depth information of the current frame; and with an NCC score voting method, all pixel depth prediction values of the current frame are selected based on two kinds of projection results to obtain a predicted depth map, and Gaussian fusion is carried out on the predicted depth map to obtain a final depth map. In addition, the invention also provides a corresponding real-time dense monocular SLAM system based on an online learning depth prediction network.

Provided is an auxiliary ultrasonic scanning system of a robot based on an RGB-D sensor. The auxiliary ultrasonic scanning system comprises a Kinect sensor, the robot, an ultrasonic probe, a marker and a host and is characterized in that the Kinect sensor serves as a visual servo system of the robot; the ultrasonic probe is clamped on a mechanical arm of the robot; the marker is fixed onto the ultrasonic probe; the visual servo system of the robot is utilized for synchronously acquiring an RGB color image and a depth image and sending images to the host; the host finishes following processes such as image stitching operation and image three-dimensional reconstruction; according to image pairs acquired by the visual servo system, the marker fixed onto the ultrasonic probe is recognized and positioned by the host; according to a recognition-based positioning result, the position and the posture of the ultrasonic probe are calculated; and the host is used for sending a control instruction to the robot so that the mechanical arm of the robot is controlled to reach at the specified position for carrying out ultrasonic scanning operation. The auxiliary ultrasonic scanning system of the robot based on the RGB-D sensor is advantaged by being reasonable in design, reliable in performance, high in automated degree and detection efficiency and low in cost.

An image processing apparatus for generating image data for displaying a stereoscopic image on the basis of an image for a left eye and an image for a right eye, the image processing apparatus includes first detecting means for detecting image pairs each taken in a continuous shooting mode among a plurality of images, second detecting means for detecting image pairs each taken consecutively within a predetermined time among the plurality of images, presenting means for presenting the image pairs detected by the first detecting means or the second detecting means to a user, and generating means for setting one of two images comprising a the image pair selected by the user among the image pairs presented by the presenting means as the image for the left eye and the other as the image for the right eye, and generating the image data for displaying the stereoscopic image.

The invention provides a multi-line array laser three-dimensional scanning system and method. Accurate synchronism and logic control of the multi-line array laser three-dimensional scanning system can be achieved through an FPGA, a line laser unit array serves as a projection pattern light source, trigger signals are sent to a stereoscopic vision image sensor and the line laser unit array through the FPGA, an upper computer receives image pairs shot by the stereoscopic vision image sensor, laser line array patterns in the image pairs are subjected to encoding, decoding and three-dimensional reconstruction, three-dimensional reconstruction and matching alignment of three-dimensional feature points between different moments are conducted on surface feature points of an object to be measured, and matching calculation is subjected to prediction and error correction through an optical tracking technology. The system and method are used for registration and connection of time domain laser three-dimensional scanning data, meanwhile, measuring error grades are evaluated in real time and fed back to an error feedback controller for adjusting indication, and therefore laser three-dimensional scanning with low cost and high efficiency, reliability and accuracy is completed accordingly.

Method and system for creating a fused image from an image pair comprising a high resolution panchromatic image and lower resolution multi-spectral image. The method includes obtaining image data (204) defining a first image of a panchromatic image type and a second image of a multi-spectral image type. The first image has a first spatial resolution and a first spectral resolution. The second image has a second spatial resolution which is lower than the first spatial resolution and a second spectral resolution higher than the first spectral resolution. The method also includes a step (212) of concurrently calculating a point-spread function for down-sampling the first image to the second spatial resolution, and a set of weights for down-sampling the second image to the first spectral resolution.

A three-dimensional helmet display of an augmented reality system is characterized by adopting an optical perspective type structure; being horizontally symmetrical by taking the nose part of a user as a central line; and being provided with an image source, a polarization azimuth adjustment mirror with adjustable azimuth, an optical imaging system and a polarization combination mirror in sequencefrom the external side to the human eye center along the optical axis direction, wherein the polarization combination mirror leads the light direction of the image source to be reflected and deflected by 90 degrees and then enter human eyes; the external light forms a 90-degree angle with the image source light, and then enters human eyes through the transmission of the polarization combination mirror after pass through the polarization mirror with the adjustable azimuth; two images forming a three-dimensional image pair are respectively transmitted to the image sources at the left side and the right side; the polarization azimuth adjustment mirror and the polarization combination mirror constitute an image source brightness adjustment unit; and the polarization mirror and the polarization combination mirror constitute an external light brightness adjustment unit. The invention can lead the virtual environment and the real environment of the three-dimensional helmet display to achievethe best matching state, eliminates the two-eye competition phenomenon, and is used for the augmented reality system.