Abstract: Providing high-quality images of members of a group moving within a coverage area that includes non-group members is provided. A current location of each member of the group is tracked within the coverage area. Images are obtained of the coverage area. Based on the tracked current location of each member of the group within the coverage area, at least one of the non-group members are selectively redacted from the images of the coverage area such that the members of the group are clearly shown in redacted images and privacy rights of the at least one of the non-group members are protected. The selectively redacted images are provided to at least one of the members of the group.

Abstract: In implementations of transferring image style to content of a digital image, an image editing system includes an encoder that extracts features from a content image and features from a style image. A whitening and color transform generates coarse features from the content and style features extracted by the encoder for one pass of encoding and decoding. Hence, the processing delay and memory requirements are low. A feature transfer module iteratively transfers style features to the coarse feature map and generates a fine feature map. The image editing system fuses the fine features with the coarse features, and a decoder generates an output image with content of the content image in a style of the style image from the fused features. Accordingly, the image editing system efficiently transfers an image style to image content in real-time, without undesirable artifacts in the output image.

Abstract: Various approaches related to capturing content of erasable boards are discussed herein.

Abstract: A method is provided. The method includes receiving at least part of an input image file. The at least part of the input image file includes input image data representing an image and reference tone mapping strength data representing a reference tone mapping strength parameter for deriving an input value representing an amount of spatially-variant tone mapping. The method further includes inputting the input image data and the input value to a tone mapping operation. The tone mapping operation is applied to the input image data to generate output image data representing an output image with the amount of spatially-variant tone mapping applied. A computing device is also provided.

Abstract: A multiple exposure method is provided. A material picture acquisition request including a search keyword is sent, by a terminal, to a server and a material picture is received in response to the material picture acquisition request. The search keyword includes geographic location information based upon which the material picture is obtained among material pictures associated with corresponding picture associated information. The picture associated information includes picture location information. The material picture is processed by the terminal based on a resolution of a currently captured picture and content of the material picture to obtain a fusion region. Mixing synthesis is performed, by the terminal, on the fusion region and the currently captured picture to obtain a mixed frame.

Abstract: An image processing apparatus comprises: an acquisition unit that acquires images of a plurality of consecutive frames; a first composite unit that performs lighten composite on the images; a second composite unit that performs darken composite on the images by two consecutive frames, and subtracts an image of a frame before or after the images of the two frames subjected to the darken composite from a result of the darken composite to generate a subtraction image; a third composite unit that performs lighten composite on subtraction images obtained by the second composite unit; and an addition unit that adds a first image obtained by performing the lighten composite on the images by the first composite unit and a second image obtained by performing lighten composite on the subtraction images.

Abstract: A method and an apparatus for generating a set of processed images of an object from overlapping raw images are provided. The raw images are aligned so that pixels representing a same part of the picture object in different raw images are aligned forming a respective pixel stack for each pictured part of the object. At least one mask image is then generated by a maximum intensity projection through the pixel stacks. The set of processed images is then generated from the raw images using the at least one mask image.

Abstract: During a period in which it is necessary to display two or more images in a superimposed manner, an image generation block of an information processing apparatus supplies the data of the image concerned to a superimposition processing block. A first image acquisition block and a second image acquisition block of the superimposition processing block acquire a first image and a second image, respectively, to be displayed in a superimposed manner. A luminance range adjustment block adjusts at least one luminance range as required with amount of adjustment corresponding to an alpha (?) value set to the second image. An output value computation block executes the computation of alpha blending by use of the image after adjustment so as to determine the pixel value of a display image, thereby outputting the determined pixel value to an image output block.

Abstract: Aspects are disclosed for enhancing digital media. In an aspect, a target object in a primary image is identified, and reference images that include the target object are located. The target object is then modified within the primary image according to data derived from analyzing the reference image. In another aspect, a primary file is received, and at least one reference file is referenced to generate enhancement data that facilitates enhancing the primary file from an extrapolation of the reference file. In yet another aspect, media files corresponding to a common event are aggregated, and a desired enhancement of a primary file is identified. Here, the desired enhancement corresponds to a modification of an obstruction included in the primary file. A reference file which includes data associated with the desired enhancement is then referenced, and the obstructed data is modified based on replacement data extrapolated from the reference file.

Abstract: A system for checking a package body with an image is capable of automatically judging good or bad product of manufactured package bodies and automatically performing change of production conditions in a filling and packaging machine, and includes a filling and packaging machine, an image data judgement system, a storage device and an image data analysis system, wherein the image data judgement system is provided with an imaging device for shooting a package body, a storage unit for storing judgement condition on good or bad package body, a judgement unit for judging good or bad package body and an image data transmission unit for transmitting the image data shot by the imaging device to the storage device, and the image data analysis system is provided with an image data acquisition unit for acquiring image data and so on judged as a bad product in the judgement unit from the storage device.

Abstract: Devices and methods for evaluating at least one photograph during a photography session are described. In some instances, a camera may be used to capture images during a photography session and a computing device may be used to identify a portrait order specification for the photography session including a set of required photographs, each having associated required criteria. The computing device may evaluate the images and determine whether the images can be associated with any of the required photographs based on features included in the images being associated with the required criteria of the required photographs. The images associated with required photographs may be displayed, and a determination of whether the images qualify as a required photograph may be made using a graphical user interface, or may be made automatically.

Abstract: A system and method for non-destructively determining characteristics of a vegetable or fruit may include processing an image of the vegetable or fruit to produce image analysis results; analyzing hyperspectral and/or Near Infrared (NIR) illumination reflected from the vegetable or fruit to produce reflection analysis results; and calculating at least one value that reflects at least one characteristic of the vegetable or fruit based on the image analysis results and based on the reflection analysis results.

Abstract: User interfaces for managing defects are provided. A defect selection interface may include a set of defect items for selection by a user. The defect selection interface may include one or more first visuals indicating similarity of the set of defect items to other defect items. An issue selection interface may include a set of issue items for selection by the user. Individual issue items may include one or more defect items added to the individual issue items. A defect comparison interface may include a comparison of a defect item to an issue item. The defect comparison interface may include one or more second visuals indicating similarity of the defect item to the issue item. Based on the user's selection, the defect item may be added to the issue item.

Abstract: An individual identifying device includes a generation unit and an imaging unit. The generation unit generates a pattern on an object. The imaging unit acquires an image of the pattern after the pattern is stabilized, in conjunction with generation of the pattern by the generation unit.

Abstract: A convolutional neural network may be trained to inspect subjects such as carbon fiber propellers for surface flaws or other damage. The convolutional neural network may be trained using images of damaged and undamaged subjects. The damaged subjects may be damaged authentically during operation or artificially by manual or automated means. Additionally, images of undamaged subjects may be synthetically altered to depict damages, and such images may be used to train the convolutional neural network. Images of damaged and undamaged subjects may be captured for training or inspection purposes by an imaging system having cameras aligned substantially perpendicular to subjects and planar light sources aligned to project light upon the subjects in a manner that minimizes shadows and specular reflections. Once the classifier is trained, patches of an image of a subject may be provided to the classifier, which may predict whether such patches depict damage to the subject.

Abstract: A method includes: receiving a defect map from a defect scanner, wherein the defect map comprises at least one defect location of a semiconductor workpiece; annotating the defect map with a reference fiducial location of the semiconductor workpiece; determining a detected fiducial location within image data of the semiconductor workpiece; determining an offset correction based on comparing the detected fiducial location with the reference fiducial location; producing a corrected defect map by applying the offset correction to the defect map, wherein the applying the offset correction translocates the at least one defect location; and transferring the corrected defect map to a defect reviewer configured to perform root cause analysis based on the corrected defect map.

Abstract: In general, techniques are described for processing a set of high-resolution images of an integrated circuit, the images captured at different locations with respect to the integrated circuit, to automatically align and “stitch” the set of high-resolution images into a larger composite image. For example, an imaging system as described herein may use sampled feature points distributed across different grid tiles within overlap regions for pairs of images to match feature points to inform the alignments of a pair with respect to each image in the pair. The system may in some cases further apply a bundle adjustment to iteratively align and refine the alignment results for each image in a set of images being processed. In some examples, the bundle adjustment is a best-fit adjustment based on minimizing the net error associated with the alignment of the set of images.

Abstract: The following concerns a method for co-localization of microscopy or histology stains by the assembly of a virtual image from one or more imaging operations. In particular, the method decreases the time required to obtain multiple labeled antigen or protein histology images of a biological sample. The method includes imaging the tissue as it is sliced by a microtome with a knife edge scanning microscope and spatially aligning the samples by the generated images. The spatial alignment of samples enabled by the method allows a panel of different antigen or protein secondary or functional stains to be compared across different sample slices, thereby allowing concurrent secondary stains of tissues and cells.

Abstract: Various arrangements for identifying and grading cancer in tissue samples are presented. A digital image of a stained tissue sample may be acquired. A Shearlet transform may be performed on the digital image of the stained tissue sample. Shearlet coefficients may be calculated based on the performed Shearlet transform of the normalized digital RGB image of the stained tissue sample. A trained neural network may be applied to create a plurality of feature maps using the digital image and Shearlet coefficients, wherein the trained neural network was trained using a plurality of images and Shearlet coefficients of a plurality of digital images. A classifier may be applied to an output of the trained neural network to identify whether cancer is present in the stained tissue sample. A notification may be output that is indicative of a grade of detected cancer in the sample.

Abstract: Provided is a similar case search device which rapidly searches for an appropriate similar case on the basis of plural ROIs. An individual similarity calculation unit sets the ROIs and plural case lesions in a case image so as to be in one-to-one correspondence with each other, compares the feature amounts of the ROIs and the feature amounts of the case lesions which correspond to each other, and calculates an individual similarity for each ROI. A total similarity calculation unit calculates a total similarity for only combinations of completely different types, which are combinations of plural ROIs and plural different types of case lesions in the same case, on the basis of plural individual similarities calculated for each of the plural ROIs. A similar case search unit searches for a similar case on the basis of the total similarities corresponding to the combinations of completely different types.

Abstract: We present a method to detect dots in microscope images of tissue samples for the purpose of diagnosis or therapy selection. Dots in tissue arise trough CISH or FISH staining, but also through staining of small cellular compartments, or in other ways. The method is applicable to both brightfield and fluorescence modalities, as well as mass spectrometry imaging methodologies. The method is based on the use of first and second derivatives of the image along the image axes, computed using regularized derivative operators, i.e. Gaussian derivatives. The outputs of these operators are examined at appropriate distances from each location in the image. The found values must have specific signs. If the signs are all correct, these values are multiplied together to obtain a confidence measure for a dot being present at that location. If the signs do not match, no dot is present, and a zero is output.

Abstract: Embodiments access a pre-immunotherapy image of tissue demonstrating NSCLC including a tumor and a peritumoral region; extract a first set of radiomic features from the image; provide the first set of radiomic features to a first machine learning classifier; receive a first probability from the first classifier that the tissue is hyperprogressor (HP) or non-responder (R); if the first probability that the tissue is within a threshold: generate a first classification of the ROT as HP or non-R based on the first probability; if the first probability is not within the threshold: extract a second set of radiomic features from the peritumoral region and provide the second set to a second machine learning classifier; receive a second probability from the second classifier that the tissue is HP or R; generate a second classification of the tissue as HP or R based on the second probability; and display the classification.

Abstract: A method of training neural networks using digital images is disclosed. Each of the digital images is associated with an imaging modality, an anatomy of a person, a diagnosis of the person, an age of the person, a gender of the person, data indicative of contouring, an aggregated score, or a property of the anatomy of the person. In the method, a client side viewing application receives the digital images, receives a modification of one of the digital images, and transmits the modified digital image to a server. The server selects a neural network to train based on the digital image, trains the selected neural network, receives a request to apply one of the neural networks to another set of digital images, and selects a neural network to apply. The neural network is applied to the other set of digital images and generates a graphical, audio, alphanumeric text, or video result.

Abstract: An imaging (e.g., with an optical coherence tomography system) method that includes 1) acquiring repeated B-scans in a manner consistent with forming images, 2) processing the acquired images according to a variable interscan time analysis (VISTA) method, and 3) generating and displaying a color-mapped image pixel color of the color-mapped image fluid flow speed, or a related quantity.

Abstract: A simulate segmentation method of cylinder and pie cake digital models utilizes a three-dimensional model and a reference point to cope with various shapes of the nuclear reactor structures. The segmentation simulation of the nuclear reactor structure is conducted with genetic algorithm. The segmentation simulation of the nuclear reactor structure is achieved by using the genetic algorithm to perform a double selection mechanism on the cross-sectional area of the nuclear reactor structure to select the optimal configuration of the segmentation, thus minimizing the cross-sectional areas of the nuclear reactor structure. The cutter segments the nuclear reactor structure based on the optimal configuration of the segmentation, thereby achieving the purpose of minimizing the attrition rate of a cutter and segmenting the nuclear reactor structure.

Abstract: An image processing apparatus and method for object segmentation in color image frames, is provided. The image-processing apparatus generates, based on a first neural network model, a first foreground mask and a second foreground mask for an object-of-interest in a first color image frame and an upcoming color image frame, respectively, of the sequence of color image frames. The image-processing apparatus determines a third foreground mask based on interpolation of the first foreground mask and the second foreground mask and updates, by use of the second neural network model, the third foreground mask to a fourth foreground mask. The image processing apparatus segments the object-of-interest from at least the first color image frame, the upcoming color image frame, and the intermediate color image frame by use of the first foreground mask, the second foreground mask, and the fourth foreground mask, respectively.

Abstract: An image edge extraction device for an automated driving system is capable of extracting an image edge in a rapid manner. The device employs an image edge extraction method that includes: acquiring a grayscale image of a target image; calculating an edge magnitude and an edge angle of the grayscale image, so as to generate an edge magnitude image and an edge angle image; detecting an edge peak of the edge magnitude image in accordance with the edge angle image, so as to acquire each edge point of the edge magnitude image and acquire an edge peak detection result; and performing double-thresholding segmentation on the edge peak detection result through a flooding method, so as to acquire an image edge of the target image.

Abstract: Techniques to improve edge detection in images. Some techniques include logic to process image data into patched image data in accordance with a colorspace model where the patched image data includes color data in a plurality of patches and identify an image group corresponding to the patched image data. The logic may be further configured to select, based upon the image group, a colorspace transform mechanism being operative to transform the image data into transformed image data in accordance with another colorspace model, the other colorspace model having a higher likelihood than the colorspace model at edge detection for the image group. The logic may be further configured to apply the colorspace transform mechanism to the image data to generate the transformed image data in accordance with the other colorspace mode and then, apply an edge detection technique to the transformed image data. Other embodiments are described and claimed.

Abstract: A system and method for using gaze information to extract visual attention information combined with computer derived local saliency information from medical images to (1) infer object and background cues from a region of interest indicated by the eye-tracking and (2) perform a medical image segmentation process. Moreover, an embodiment is configured to notify a medical professional of overlooked regions on medical images and/or train the medical professional to review regions that he/she often overlooks.

Abstract: An image processing apparatus specifies, in a foreground region which includes a foreground object in an input image obtained by image capturing, a first region including a predetermined target object, and outputs an output image generated by abstracting the first region without abstracting at least a part of a second region which does not include the predetermined object in the foreground region.

Abstract: A vehicle that includes one or more sensors configured to output data and an electronic control unit. The electronic control unit is configured to receive first data from the one or more sensors at a first time, detect an object based on the first data received from the one or more sensors at the first time, classify the object into a classification based on the first data received from the one or more sensors at the first time, and receive second data from the one or more sensors at a second time. The second time is subsequent to the first time. The electronic control unit is further configured to select a proper subset of the second data for further processing based on the classification of the object, and track the object based on the proper subset of the second data selected for further processing.

Abstract: According to an aspect, a method of position-based adjustment to display content is provided. The method includes determining a position of an observer relative to a display device. The method also includes determining a distortion correction to apply to a plurality of display content based on the position of the observer relative to the display device to correct the display content with respect to the observer. The distortion correction of the display content is output to the display device.

Abstract: Methods and systems for enhanced object tracking by receiving sensor fusion data related to target objects and object tracks; determining splines representing trajectories of each target object; filtering the sensor fusion data about each target object based on a first, second and third filtering model wherein each filtering model corresponds to one or more of a set of hypotheses used for processing vectors related to trajectories of a track object wherein the set of hypotheses comprise: a path constraint, a path unconstrained, and a stationary hypothesis; and generating a hypothesis probability for determining whether to use a particular hypothesis based wherein the hypothesis probability is determined based on results from the first, second and third filtering models and from results from classifying, by at least one classification model, one or more features related to the object track for the target object.

Abstract: An information processing apparatus includes a memory configured to store a recognition result, accuracy and timing of an image captured at a point, and a processor coupled to the memory and the processor configured to specify the image configured to include a designated recognition object, and search for the image having a same recognition result as the recognition object by referring to the memory, based on the point and timing at which the specified image has been captured and movement speed information of the recognition object.

Abstract: An information processing device includes: a memory; and a processor coupled to the memory and configured to: estimate a first position of a first joint in a two-dimensional plane based on a color image, and estimate a first positional relationship between the first joint and a second joint coupled to the first joint via a portion in the two-dimensional plane; estimate a second position of the first joint in the two-dimensional plane based on a depth image indicating respective depths of the first joint and the second joint, and estimate a second positional relationship between the first joint and the second joint in a three-dimensional space; and estimate a third position of the first joint in the two-dimensional plane and posture of the portion in the three-dimensional space based on the first position, the first positional relationship the second position, and the second positional relationship.

Abstract: A method of measuring infrared spectral characteristics of a moving target, the method including: establishing a multi-dimensional and multi-scale model with infrared spectral features of an object-space target, and extracting an object-space region of interest measurement model; performing target detection on an actually measured infrared image, and identifying position information for each ROI of a target; tracking the target, to obtain the target's pixel differences between two frames, as well as a moving direction of the target, and performing motion compensation for the target; and scanning the target, and after successfully capturing an image of the target being tracked, controlling an inner framework to point to each target of interest, and according to moving-direction information of the target, performing N-pixel-size motion in a direction shifted by 90° with respect to the moving direction, and activating a spectrum measuring module.

Abstract: Methods and systems integrate digital fingerprint authentication-based identification and location tracking into a single, continuous process in which an authentication-integrated tracking system is simultaneously aware of both the identity and location of each physical object at all times as they move along a conveyance system. Insertion or removal of an object is quickly detected and reported. Rapid reestablishment and continuation of authentication-integrated tracking is enabled in the event of any temporary interruption or failure of tracking in the system. An exemplary system comprises plural tracking units networked together, each tracking unit including a camera or scanner to observe a corresponding field of view, the tracking units arrange to realize a continuous field of view of a physical conveyance system.

Abstract: An image processing apparatus including an information acquisition unit that acquires information regarding a camera, an estimation unit that estimates an object of interest during shooting, in an image captured by the camera, based on the information regarding the camera, an area dividing unit that divides the image into a plurality of divided areas, using each of a plurality of types of dividing methods, and a motion vector detection unit that detects a motion vector in the plurality of divided areas regarding each of the plurality of types of dividing methods. The area dividing unit changes at least one of a size of a divided area and a number of division in each of the plurality of types of dividing methods for dividing the image, according to an object of interest estimated by the estimation unit.

Abstract: Systems and methods for moving point detection for point clouds. Some implementations may include obtaining a sequence of point clouds, wherein the sequence of point clouds includes a current point cloud, a previous point cloud that precedes the current point cloud, and a next point cloud that follows the current point cloud; inputting a voxelized representation of the current point cloud to a forward-pass recurrent neural network with a hidden state based on the previous point cloud to obtain a first voxelized activation; inputting the voxelized representation of the current point cloud to a backward-pass recurrent neural network with a hidden state based on the next point cloud to obtain a second voxelized activation; combining the first voxelized activation and the second voxelized activation to obtain a prediction of whether respective points are moving or static; and updating the current point cloud based on the prediction.

Abstract: An electronic device for object tracking based on a user-specified initialization point is provided. The electronic device stores a sequence of image frames, which includes a first image frame and a second image frame. The electronic device estimates a set of feature correspondences between a first set of features points in the first image frame and a second set of feature points in the second image frame. The electronic device generates different first motion-estimate models for different groups of feature correspondences of the set of feature correspondences and further estimates, from different groups of feature correspondences, a plurality of inlier feature correspondences that correspond to the object of interest in the first image frame and the second image frame. The electronic device generates a second motion-estimate model as an optimal motion-estimate model and tracks the object of interest in the sequence of frames, based on the second motion-estimate model.

Abstract: An information processing apparatus includes a processor that detects positions of ball candidates from a plurality of time-series image frames. The processor adds a position of a second ball candidate to a second image frame subsequent to a first image frame based on a position of a first ball candidate and movement definition information. The first ball candidate is detected from the first image frame. The movement definition information defines a characteristic of a movement of a ball. The processor generates a plurality of trajectory candidates by combining a plurality of ball candidates detected from image frames of different times. The processor evaluates the plurality of trajectory candidates to determine a ball trajectory. The processor interpolates, when the ball trajectory is interrupted, between a starting point and an ending point of the interruption with a trajectory of a first person who moves from the starting point to the ending point.

Abstract: A system and method of processing video information received from surveillance cameras for tracking moving objects in real time or archival video. The system can contain a set of image capture devices, a storage device, an image-display device, a data input device, and a set of data processing devices configured for performance of stages. The stages can include a linking of a set of image capture devices to a location map, displaying of the video image from the first image capture device, selecting a moving object which can be tracked by operator as an object of interest, identifying the movement of the selected moving object and its highlight on the video image as area of interest, assessing a direction of the movement of the selected moving object, and predicting a location of the object in the field of view of the second capture device.

Abstract: Some embodiments of an image processing apparatus comprise a processor to execute instructions. The instructions are for detecting feature points from images in a first plurality of images that at least partially overlap in angle of view and have different focus positions and from at least one second image, calculating a first conversion coefficient from the feature points, combining the images in the first plurality of images based on the first conversion coefficient, calculating a second conversion coefficient of at least a part of the images in the first plurality of images by using the feature points detected from the second image, and combining the at least a part of the images in the first plurality of images by using the second conversion coefficient, wherein a depth of field in the second image is deeper than a respective depth of field in each image in the first plurality of images.

Abstract: A method for providing depth map information based on image data descriptive of a scene. In one embodiment, after generating an initial sequence of disparity map data, performing a smoothing operation or an interpolation to remove artifact introduced in the disparity map data as a result of segmenting the image data into superpixels.

Abstract: In one embodiment, a system includes a first projector configured to project a first projected pattern having one or more first projected lighting characteristics, a second projector configured to project a second projected pattern having one or more second projected lighting characteristics, a camera configured to capture an image comprising first and second detected patterns corresponding to reflections of the first and second projected patterns, respectively, and one or more processors configured to: identify a detected point in the image that corresponds to a projected point in at least one of the first and second projected patterns by comparing detected lighting characteristics of the first and second detected patterns with the first and second projected lighting characteristics, and compute a depth associated with the detected point based on the projected point, the detected point, and a relative position between the camera and at least one of the projectors.

Abstract: A camera module with a fixed near field focus is configured to capture a single image. That single image is segmented by an image divider a number of regions. A focus metric determiner then determines a focus metric for each of the regions. A depth map generator maps the focus metric into a depth value for each of the regions and combines the depth values to generate a depth map.

Abstract: An image processing system avoids at its best restrictions in determining applicable facilities. The image processing system includes a camera for photographing a workpiece, an illumination device having light transmittance and disposed between the camera and the workpiece, and a control device for controlling the camera and the illumination device. The illumination device can change illumination patterns according to instructions from the control device. The control device controls the camera so that photographs are taken under each illumination pattern.

Abstract: An electronic device estimates a depth map of an environment based on stereo depth images captured by depth cameras having exposure times that are offset from each other in conjunction with illuminators pulsing illumination patterns into the environment. A processor of the electronic device matches small sections of the depth images from the cameras to each other and to corresponding patches of immediately preceding depth images (e.g., a spatio-temporal image patch “cube”). The processor computes a matching cost for each spatio-temporal image patch cube by converting each spatio-temporal image patch into binary codes and defining a cost function between two stereo image patches as the difference between the binary codes. The processor minimizes the matching cost to generate a disparity map, and optimizes the disparity map by rejecting outliers using a decision tree with learned pixel offsets and refining subpixels to generate a depth map of the environment.

Abstract: Disclosed herein are an apparatus and a method for generating an intermediate view image of a stereoscopic image. The apparatus may include a feature point detector detecting contours in left and right view images and detecting feature points in the contours, a corresponding point detector detecting corresponding points corresponding to the feature points of the left and right view images, and a composer generating an intermediate view image based on disparity information between the feature points and the corresponding points.

Abstract: A method includes capturing a first image and a second image of a scene using at least one imaging camera of an imaging system. The first image and the second image form a stereo image pair and each comprises a plurality of pixels. Each of the plurality of pixels in the second image is initialized with a disparity hypothesis. Matching costs of the disparity hypothesis for each of the plurality of pixels in the second image are recursively determined, from an image tile of a smaller pixel size to an image tile of a larger pixel size, to generate an initial tiled disparity map including a plurality of image tiles. After refining the disparity value estimate of each image tile and including a slant hypothesis, a final disparity estimate for each pixel of the image is generated.