Abstract: Image denoising techniques are described. In one or more implementations, a denoising result is computed by a computing device for a patch of an image. One or more partitions are located by the computing device that correspond to the denoising result and a denoising operator is obtained by the computing device that corresponds to the located one or more partitions. The obtained denoising operator is applied by the computing device to the image.

Abstract: A character recognition apparatus includes an extracting unit that extracts a numerical value reading region in which a numerical value that is a character string of a numeral is read from an image that is a character recognition target, a character recognizing unit that performs character recognition on the extracted numerical value reading region and obtains plural recognition candidates for each numerical value, and a selecting unit that selects a potential candidate from the plural recognition candidates on the basis of a numerical value range set for each numerical value reading region.

Abstract: A safety monitoring system for human-machine symbiosis is provided, including a spatial image capturing unit, an image recognition unit, a human-robot-interaction safety monitoring unit, and a process monitoring unit. The spatial image capturing unit, disposed in a working area, acquires at least two skeleton images. The image recognition unit generates at least two spatial gesture images corresponding to the at least two skeleton images, based on information of changes in position of the at least two skeleton images with respect to time. The human-robot-interaction safety monitoring unit generates a gesture distribution based on the at least two spatial gesture images and a safety distance. The process monitoring unit determines whether the gesture distribution meets a safety criterion.

Abstract: Various embodiments are disclosed for image editing. A frame is obtained from a frame sequence depicting at least one individual, and facial characteristics in the frame are analyzed. A utilization score is assigned to the frame based on the detected facial characteristics, and a determination of whether to utilize the frame is made based on the utilization score. A completeness value is assigned, and a determination is made based on the completeness value of whether to repeat the steps above for an additional frame in the frame sequence based on the completeness value. Regions from the frames are combined to generate a composite image.

Abstract: Generally discussed herein are systems and apparatuses that relate to systems and methods for analysis of 3D light radar (LiDAR) point clouds. The disclosure also includes techniques of bare earth and feature extraction. According to an example, bare earth and feature extraction can include estimation of a ground layer, generation of an inverse Above Ground Layer (AGL), identification of paths, roads, and cleared areas, identification and removal of buildings, and identification of a Human Activity Layer (HAL). This combination of features may provide for a much higher probability of detection of human activity.

Abstract: A method determining user liveness is provided that includes calculating, by a device, eye openness measures for a frame included in captured authentication data, and storing the eye openness measures in a buffer of the device. Moreover the method includes calculating confidence scores from the eye openness measures stored in the buffer, and detecting an eye blink when a maximum confidence score is greater than a threshold score.

Abstract: Certain embodiments of the invention relate to a computer-implemented method that comprises analyzing an image over a plurality of different scales, where the analyzing includes determining spectral characteristics of the image at each of the plurality of different scales, determining spatial characteristics of the image at each of the plurality of different scales, and determining a segmentation pattern (i.e., superpixel) for the image at each of the plurality of different scales based on the spectral and spatial characteristics of that particular scale. The method further includes identifying objects in the image based on portions of the segmentation patterns that are scale-invariant over the plurality of different scales. In some cases, the method can include determining statistical, textural, and/or intensity characteristics of the image at each scale, where determining the segmentation patterns for the image at each scale is further based on one or more of these characteristics.

Abstract: Systems and methods in accordance with embodiments of the invention automatically crop images based upon the location of one or more regions of interest (ROIs) identified within the cropped image. In one embodiment, an image cropping application configures a processor to utilize at least one object detector to identify at least one region of interest within an image. A plurality of candidate cropped images within the image are selected and for each candidate cropped image, an output is generated by weighting each of the at least one region of interest contained within the candidate cropped image as a function of its position within the candidate cropped image. A crop location can then be selected based upon the candidate cropped image having the largest output.

Abstract: A doze detection method, which accurately detects a blink burst and improves speed and accuracy of doze detection, includes measuring a state where the eye is substantially open as an open eye time and another state as a closed eye time, defining a time shorter than an average blink interval of a healthy adult in an alert state as a first threshold time; defining a time longer than an average closed eye time of a healthy adult in an alert state as a second threshold time; and defining blinks as a blink burst when detecting an eye opening equal to or shorter than the first threshold time. A doze state is determined when the closed eye time of a blink among the blinks during the blink burst reaches at least the second threshold time, the blink occurring after an open eye time equal to at most the first threshold time.

Abstract: An apparatus for automatic selection of optimal tomography slices by executing a number of tentative segmentations using the same interactively provided in-slice seed point on some or all available tomography slices. The appropriate segmentation boundaries are then marked and the slice with the best segmentation goodness value (figure of merit) is presented to the user via a viewer. The steps are repeated when the user changes the seed point. The optimal segmentation boundary is displayed on top of a single simulated mammography image, fused from all tomography slice images.

Abstract: This disclosure provides a method, system and computer program product for denoising an image by extending a Block Matching and 3D Filtering algorithm to include decomposition of high contrast image blocks into multiple layers that are collaboratively filtered. According to an exemplary method, the high contrast image blocks are decomposed into a top layer, a bottom layer and a mask layer.

Abstract: A method and apparatus for reconstructing a low-resolution image is provided. The method includes magnifying a low-resolution image; extracting a high frequency component and a texture component from the pre-magnification low-resolution image; and reconstructing an image by synthesizing the extracted high frequency component and texture component with the magnified low-resolution image. The method and apparatus synthesizes only image components without a separate external DB, making it possible to enhance sharpness of the outline of images and thus effectively reconstruct fine texture components.

Abstract: A computerized method for model-less segmentation and registration of ultrasound (US) with computed tomography (CT) images of an organ with a fluid filled chamber. The method is based on correlating between the US image(s) and the CT image(s) by processing the US image(s) by iteratively expanding the CT image segment so that the expanded CT image segment is correlated with the visual boundaries of the US image segment; transforming the CT image(s) according to an estimated US transducer position and estimated US beam direction related to the US image(s) so that at least one of shape and volume of the organ in the CT image is adapted with at least one of shape and volume of the organ of the US image, to form a CT image representation which is correlated with US image(s).

Abstract: A method for determining user liveness is provided that includes extracting, by a processor, overlapping differential signals from a first differential signal. Moreover, the method includes calculating principal component analysis coefficients for each extracted differential signal, selecting a subset of the principal component analysis coefficients for each extracted differential signal, and generating an activity result for each extracted differential signal based on the principal component analysis coefficient subset.

Abstract: An image processing apparatus includes: a processor coupled to a memory, configured to: perform an analysis of resolution in at least two directions of an image which is taken from a subject having a radial pattern, and determine filter data containing an adjusted weight coefficient which is obtained by adjusting a weight coefficient in one of the two directions, whichever has lower resolution, based on a result of the analysis of the image corrected by filtering on the image in accordance with a blurring function of the image.

Abstract: A hierarchy of clusters is determined, where each leave of the hierarchy corresponds to one of the images in a group, and each cluster in the hierarchy identifies images in the group that are deemed similar to one another. The hierarchy identifies a similarity between each of the plurality of clusters.

Abstract: An inspection apparatus inspects a printed product by positioning a reading target image of the printed product relative to a reference image and collating the reading target image with the reference image includes a positioning unit configured to perform positioning processing for the reference image and the reading target image with a first precision, and a detection unit configured to detect an image defect candidate area by collating the reading target image with the reference image, which have been positioned by the positioning unit. The positioning unit newly performs the positioning processing for the reference image and the reading target image in the image defect candidate area with a second precision that is higher than the first precision. The detection unit detects an image defect by collating the reading target image with the reference image in an area having been newly subjected to the positioning processing using the second precision.

Abstract: An image processing apparatus includes a calculation section configured to calculate filtering coefficients of a filter with a first area in an image that is partitioned into multiple first areas including the first area, the image being partitioned differently into multiple second areas, each one of which being covered by several first areas, to calculate a convoluted image of a second area using the filtering coefficients calculated with the first areas covering a part of the second area, the calculation being executed for the several first areas covering distinct parts of the second area, respectively; and an interpolation section configured to interpolate a pixel in the second area using pixels at the same position in the convoluted images of the second area which are convoluted with the respective filtering coefficients.

Abstract: The present disclosure proposes a method of moving object detection in variable bit-rate video steams based on probabilistic neural networks, and the method features a background generation module and a moving object detection module. The background generation module produces a model of background images which express properties of variable bit-rate video streams. The moving object detection module distinguishes a moving object in both low and high bit-rate video steams in an efficient manner. The detection result is generated by calculating the output value of the probabilistic neural networks.

Abstract: A detection learning module is used for enabling and/or disabling real-time compression detection by maintaining a history of real-time compression detection success for sampled data. The enabling or disabling of the real-time compression detection is based on a detection benefit function derived from a set of calculated heuristics indicating the real-time compression detection success on input streams.