Abstract: This image quality evaluation device comprises: a conversion unit that converts the data of an image to two-dimensional array data of the luminance value of each pixel in the image; a processing unit that executes an averaging process, based on a plurality of filter sizes, on each pixel of the two-dimensional array data; and an evaluation unit that evaluates the quality of the image using the processing results of the processing unit.

Abstract: A method for measuring volume of pomelo based on three view image contour fitting is provided, for measuring the volume of the pomelo based on images, includes: a number of image capturing devices capturing a number of surface images of a pomelo to be measured; calibrating a position of an image capturing device for capturing a top view image of the pomelo to be measured; performing enhanced extraction processing on the number of surface images, and obtaining a standard contour image; obtaining a size of each pomelo slice based on the standard contour image, and obtaining a B-spline curve by fitting a contour of the pomelo slice; and calculating volume of each pomelo slice according to the B-spline curve, and obtaining total volume of the pomelo to be measured by calculating a sum of the volume of all of the pomelo slices.

Abstract: A computing system including a processing circuit in communication with a camera having a field of view. The processing circuit is configured to perform operations related to detecting, identifying, and retrieving objects disposed amongst a plurality of objects. The processing circuit may be configured to perform operations related to object recognition template generation, feature generation, hypothesis generation, hypothesis refinement, and hypothesis validation.

Abstract: An aspect of the present invention is a learning method for learning a model which reconstructs data and acquires reconstructed data, the learning method includes a plurality of stages, and the learning method includes first sub-processing which is based on an iterative calculation algorithm in which a Lagrange multiplier is not used, second sub-processing which is based on an iterative calculation algorithm in which the Lagrange multiplier is used, and weighting sub-processing which determines processing in each of the plurality of stages based on a weight for determining usage of the first sub-processing and the second sub-processing in each stage.

Abstract: The present disclosure is directed toward systems, methods, and non-transitory computer readable media that provide self-supervised object discovery systems that combine motion and appearance information to generate segmentation masks from a digital image or digital video and delineate one or more salient objects within the digital image/digital video. The disclosed systems utilize a neural network encoder to generate a fully connected graph based on image patches from the digital input, incorporating image patch feature and optical flow patch feature similarities to produce edge weights. The disclosed systems partition the generated graph to produce a segmentation mask. Furthermore, the disclosed systems iteratively train a segmentation network based on the segmentation mask as a pseudo-ground truth via a bootstrapped, self-training process.

Abstract: A traffic light recognizing apparatus includes: an imager configured to capture a color image of driving environment; a lighting region detector detecting a lighting region in a traffic light body on the image; a first lamp detector detecting a lamp in an ON state in the traffic light body based on the lighting region; a first search area setter setting search areas having the same dimensions as the lamp and horizontally and vertically aligned with the lamp on the image, in accordance with a lighting color of the lighting region; a second lamp detector evaluating luminances at an edge and in a center of each search area, and detecting lamps in an OFF state in the traffic light body; and an orientation determiner determining whether the traffic light body is horizontally or vertically disposed based on the arrangement of the lamps in the ON and OFF states.

Abstract: The present disclosure relates to a method of inferring an epipolar plane image and extrapolating a view of a target image using the epipolar plane image. A view extrapolation method using an epipolar plane image according to an embodiment of the present disclosure includes: creating an Epipolar Plane Image (EPI) using a target image and a depth map corresponding to the target image; creating a super-resolution EPI and a disparity of the EPI on the basis of the EPI; creating an EPI mask by warping the super-resolution EPI in accordance with the disparity; and creating a restored EPI by applying the EPI mask to the warped super-resolution EPI, and creating an extrapolated image for the target image using the restored EPI.

Abstract: An image processing device configured to convert image data of a first resolution into image data of a second resolution higher than the first resolution includes at least one memory, and at least one processor in communication with the at least one memory and configured to cooperate with the at least one memory to calculate a direction and an intensity of an edge from the image data of the first resolution, and determine a pattern for the image data of the second resolution to be replaced by pixels of the image data of the first resolution, based on the direction and the intensity of the edge.

Abstract: A system for detecting an event occurrence period of cyclical event includes an image monitoring device and a period judgment device. The image monitoring device captures monitoring images respective in image capturing times. The period judgment device extracts feature vectors from the monitoring images to calculate vector values to accordingly generate a vector value time domain signal, executes a short-time Fourier transform (STFT) with a window width time to transform the vector value time domain signal to a spectrogram, makes the spectrogram be denoised to generate a representative frequency time domain signal, and calculates an event-cycle achievement rate and a product term accumulation number after each window width time, so as to accordingly calculate event occurrence time and the event occurrence period.

Abstract: A system and corresponding method perform image super-resolution (SR). The system comprises an element-unshuffled downsampler and an image SR module. The image SR module performs image SR on a low-resolution (LR) representation of a high-resolution (HR) original image. The HR original image is at a higher resolution relative to a resolution of the LR representation. The image SR module produces a reconstructed version of the HR original image via the image SR performed. The image SR is based on element-unshuffled downsampling of the LR representation. The element-unshuffled downsampler performs the element-unshuffled downsampling. The image SR module outputs the reconstructed version produced. The system performs the image SR with fewer parameters and less computation cost relative to conventional image SR.

Abstract: An image deblurring method and apparatus are provided. The image deblurring method includes generating a primary feature representation on a first blur point in an input image and offset information on similar points of the first blur point by encoding the input image by implementing an encoding model, generating secondary feature representations on the similar points by applying the offset information to the primary feature representation, and generating an output image, based on the secondary feature representations and the offset information, by implementing an implicit function model.

Abstract: According to embodiments, an inspection apparatus includes an imaging mechanism, an image acquisition circuit that extracts an outline from image data of a sample, a development circuit that generates a developed image, an outline data generation circuit that generates data of an outline point of a pattern of the developed image, an area calculation circuit that calculates an area of a region not included in the pattern in a circle centered on the outline point, an estimation circuit that calculates a resizing amount of the outline point based on the area, and a reference image generation circuit that executes a resizing process of data of the outline point based on the resizing amount and generates a reference image based on the data of the outline point subjected to the resizing process.

Abstract: A method and system for generating a depth map corresponding to a frame of a sequence of frames in a video clip is disclosed. This can involve generating a single image depth map for each of a plurality of frames, scaling the single image depth maps, and processing a time sequence of scaled single image depth maps to generate said depth map corresponding to the frame of the sequence of frames in the video clip.

Abstract: A method includes: obtaining a first image including a foreground, a second image including a background, and soft segmentation of the foreground; and inputting the first image, the second image, and the soft segmentation into an image matting network, and outputting a foreground segmentation of the first image. The image matting network includes at least one stage network that includes a context combining module, a stem block, and a predictor module. The first image, the second image, and the soft segmentation of the foreground are inputted into the context combining module, the context combining module is configured to output a low-order feature and a high-order feature after feature exchange, the stem block is configured to fuse the low-order feature and the high-order feature based on an attention mechanism to obtain a fused feature, and the predictor module is configured to output the foreground segmentation according to the fused feature.

Abstract: An Augmented Reality (AR) system is provided. The AR system uses a combination of gesture and DMVO methodologies to provide for the user's selection and modification of virtual object of an AR experience. The user indicates that they want to interact with a virtual object of the AR experience by moving their hand to overlap the virtual object. While keeping their hand in an overlapping position, the user rotates their wrist and the virtual object is rotated as well. To end the interaction, the user moves their hand such that their hand is no longer overlapping the virtual object.

Abstract: A method for training a convolutional neural network. For each of a multiplicity of training input images, the method includes processing of the training input image by the convolutional network; processing of a scaled version of the training input image by the convolutional network; determining a pair of convolutional layers of the convolutional network so that a convolutional layer of the pair generates a first feature map for the training input image which has the same size as a second feature map which is generated by the other convolutional layer of the pair for the scaled version of the training input image; and calculating a loss between the first feature map and the second feature map; and training the convolutional neural network to reduce an overall loss which includes the calculated losses.

Abstract: Image data representing one or more images at a first resolution is received at a first artificial neural network (ANN). The image data is processed using the first ANN to generate upscaled image data representing the one or more images at a second, higher resolution. The first ANN is trained to perform image upscaling and is trained using first training image data representing one or more training images at the first resolution, the first training image data being at a first level of quality. The first ANN is also trained using features of a second ANN, wherein the second ANN is trained to perform image upscaling and is trained using second training image data representing one or more training images at the first resolution, the second training image data being at a second level of quality, higher than the first level of quality.

Abstract: A method for aligning multiple depth cameras in an environment based on image data can include accessing, by one or more processors, a plurality of first point cloud data points corresponding to a first pose relative to a subject and a plurality of second point cloud data points corresponding to a second pose relative to the subject. The method can include determining, by the one or more processors, a frame of reference for image data based on at least one of the first pose or the second pose. The method can include transforming, by the one or more processors, at least one of the plurality of first point cloud data points or the plurality of second point cloud data points to align with the frame of reference.

Abstract: Image super-resolution (SR) refers to the process of recovering high-resolution (HR) images from low-resolution (LR) inputs. Blind image SR is a more challenging task which involves unknown blurring kernels and characterizes the degradation process from HR to LR. In the present disclosure, embodiments of a variational autoencoder (VAE) are leveraged to train a kernel autoencoder for more accurate degradation representation and more efficient kernel estimation. In one or more embodiments, a kernel-agnostic loss is used to learn more robust kernel features in the latent space from LR inputs without using ground-truth kernel references. In addition, attention-based adaptive pooling is introduced to improve kernel estimation accuracy, and spatially non-uniform kernel features are passed into SR restoration resulting in additional kernel estimation error tolerance.

Abstract: A method for generating, from an input image, an output image that a given image classifier classifies into a target class chosen from multiple available classes of a given classification. The method includes mapping, using a trained encoder network, the input image to a lower dimensional representation in a latent space; drawing a noise sample from a given distribution; and mapping, using a trained generator network, the noise sample to an output image, wherein this mapping is conditioned both on the target class and on the representation.