Abstract: Frame packing techniques are disclosed for multi-directional images and video. According to an embodiment, a multi-directional source image is reformatted into a format in which image data from opposing fields of view are represented in respective regions of the packed image as flat image content. Image data from a multi-directional field of view of the source image between the opposing fields of view are represented in another region of the packed image as equirectangular image content. It is expected that use of the formatted frame will lead to coding efficiencies when the formatted image is processed by predictive video coding techniques and the like.

Abstract: Embodiments herein describe a framework for classifying images. In some embodiments, it is determined whether an image includes synthetic image content. If it does, characteristics of the image are analyzed to determine if the image includes characteristics particular to panoramic images (e.g., possess a threshold equivalency of pixel values among the top and/or bottom boundaries of the image, or a difference between summed pixel values of the pixels comprising the right vertical boundary of the image and summed pixel values of the pixels comprising the left vertical boundary of the image being less than or equal to a threshold value). If the image includes characteristics particular to panoramic images, the image is classified as a synthetic panoramic image. If the image is determined to not include synthetic image content, a neural network is applied to the image and the image is classified as one of non-synthetic panoramic or non-synthetic non-panoramic.

Abstract: Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training an optical flow object localization system and a novel object localization system. In a first aspect, the optical flow object localization system is trained to process an optical flow image to generate object localization data defining locations of objects depicted in a video frame corresponding to the optical flow image. In a second aspect, a novel object localization system is trained to process a video frame to generate object localization data defining locations of novel objects depicted in the video frame.

Abstract: One variation of a method for predicting manufacturing defects includes: accessing a first set of inspection images of a first set of assembly units recorded by an optical inspection station over a first period of time; generating a first set of vectors representing features extracted from the first set of inspection images; grouping neighboring vectors in a multi-dimensional feature space into a set of vector groups; accessing a second inspection image of a second assembly recorded by the optical inspection station at a second time succeeding the first period of time; detecting a second set of features in the second inspection image; generating a second vector representing the second set of features in the multi-dimensional feature space; and, in response to the second vector deviating from the set of vector groups by more than a threshold difference, flagging the second assembly unit.

Abstract: Various embodiments herein each include at least one of systems, methods, software, and data structures for context-aided machine vision. For example, one method embodiment includes identifying a customer in a shopping area and maintaining an item bin in a computing system of data identifying items the customer has picked up for purchase. This method further includes receiving an image of the customer holding an item and performing item identification processing on the image to identify the item the customer is holding. The item identification processing may be performed based in part on a stored shopping history of the customer indicating items the customer is more likely to purchase. The identified item is then added to the item bin of the customer.

Abstract: A method of pillow customization includes analyzing shapes associated with people through the use of sensors to create analytical data; receiving photos from a subject user through a first computing device and a server; determining a firmness of a mattress of the subject user based on the photos; determining body measurements of the subject based on the photos through one or more algorithms and a second computing device; providing the subject user with a pillow diagram, the pillow diagram having one or more zones, each of the one or more zones being customizable in firmness; receiving one or more subject user inputted selections through the first computing device; and designing a pillow based on the firmness of the mattress, the body measurements, and the one or more subject user inputted characteristics, the pillow being customized to the subject user.

Abstract: An image processor includes: an image sensor outputting a short exposure image and a long exposure image; a sensor controller that, when brightness of the subject changes, controls first exposure sensitivity to cause the short exposure image to have first brightness and controls second exposure sensitivity to cause the long exposure image to have second brightness; a motion blending ratio calculator calculating a motion blending ratio based on a motion amount of the subject; a motion-adapted image synthesizer generating a motion-adapted image by synthesizing a corrected short exposure image and the long exposure image based on the motion blending ratio; and an HDR image synthesizer generating an HDR image by synthesizing the motion-adapted image and the short exposure image together. When the subject becomes darker, the sensor controller controls the first and second exposure sensitivities to cause the first sensor gain to be at most the second sensor gain.

Abstract: A system for determining a refractive error from a red reflex image of the eyes of a subject by analyzing the red reflex image of the subject using deep learning and clustering algorithms is provided. The system (i) predicts a probability map for each of sixteen eye landmarks on a contour of an iris of the eye, (ii) determines a location of an eye landmark on the eye-region image, (iii) determines a circular mask of the iris, (iv) extracts an iris region image between the circular mask of the iris and the eye-region image, (v) generates a plurality of clusters by identifying a similarity between pixels of the iris region image, (vi) determines a crescent of the red reflex in the iris, (viii) predicts a pupil radius, (ix) determines a width of the crescent, and (x) determines a power in each eye by computing a width of an anti-crescent, eccentricity of an image capturing device, a working distance of the image capturing device.

Abstract: In implementations of skin tone assisted digital image color matching, a device implements a color editing system, which includes a facial detection module to detect faces in an input image and in a reference image, and includes a skin tone model to determine a skin tone value reflective of a skin tone of each of the faces. A color matching module can be implemented to group the faces into one or more face groups based on the skin tone value of each of the faces, match a face group pair as an input image face group paired with a reference image face group, and generate a modified image from the input image based on color features of the reference image, the color features including face skin tones of the respective faces in the face group pair as part of the color features applied to modify the input image.

Abstract: Estimating a three-dimensional (3D) pose of an object, such as a hand or body (human, animal, robot, etc.), from a 2D image is necessary for human-computer interaction. A hand pose can be represented by a set of points in 3D space, called keypoints. Two coordinates (x,y) represent spatial displacement and a third coordinate represents a depth of every point with respect to the camera. A monocular camera is used to capture an image of the 3D pose, but does not capture depth information. A neural network architecture is configured to generate a depth value for each keypoint in the captured image, even when portions of the pose are occluded, or the orientation of the object is ambiguous. Generation of the depth values enables estimation of the 3D pose of the object.

Abstract: Systems and methods are described herein for tagging images for placement in a graphical layout based on characteristics of depicted faces. The brightness of each face depicted is determined and, if the brightness of any face is below a threshold level of brightness, the image is tagged with a negative identifier indicating that the image cannot be dimmed. If the brightness of every face exceeds the threshold level of brightness, then the image is tagged with a positive identifier indicating that the image can be dimmed. The resolution of the image is also determined and, if the resolution currently affects any face, the image is tagged with a negative identifier indicating that the image cannot be blurred. If the resolution does not affect any face, and further blurring will not affect any image, then the image is tagged with a positive identifier indicating that the image can be blurred.

Abstract: Methods, systems and computer program products for flagging abnormal videos are provided. Aspects include training an image recognition model based on a plurality of images that depict one or more of a plurality of subjects. Aspects also include generating a normal subject relationship graph representing normal relationships between the plurality of subjects by applying the image recognition model to a plurality of training videos and a test subject relationship graph representing test relationships between subjects depicted in a test video by applying the image recognition model to the test video. Each normal relationship is associated with a strength value. Responsive to determining that a difference between a strength value associated with a first normal relationship and a strength value associated with a corresponding first test relationship exceeds a predetermined threshold, aspects include flagging the test video as being abnormal.

Abstract: Described is a system for converting a convolutional neural network (CNN) designed and trained for color (RGB) images to one that works on infrared (IR) or grayscale images. The converted CNN comprises a series of convolution layers of neurons arranged in a set kernels having corresponding depth slices. The converted CNN is used for performing object detection. A mechanical component of an autonomous device is controlled based on the object detection.

Abstract: A sensing and computing system and method for capturing images and data regarding an object and calculating one or more parameters regarding the object using an internal, integrated CPU/GPU. The system comprises an imaging system, including a depth imaging system, color camera, and light source, that capture images of the object and sends data or signals relating to the images to the CPU/GPU, which performs calculations based on those signals/data according to pre-programmed algorithms to determine the parameters. The CPU/GPU and imaging system are contained within a protective housing. The CPU/GPU transmits information regarding the parameters, rather than raw data/signals, to one or more external devices to perform tasks in an industrial environment related to the object imaged.

Abstract: A fingerprint verification method and apparatus is disclosed. The fingerprint verification method may include obtaining an input fingerprint image, determining a matching region between the input fingerprint image and a registered fingerprint image, determining a similarity corresponding to the matching region, representing a determined indication of similarities between the input fingerprint image and the registered fingerprint image, relating the determined similarity to the matching region as represented in a matching region-based similarity, determining a result of a verification of the input fingerprint image based on the matching region-based similarity, and indicating the result of the verification.

Abstract: Disclosed are methods, systems, and non-transitory computer-readable medium for color and pattern analysis of images including wearable items. For example, a method may include receiving an image depicting a wearable item, identifying the wearable item within the image by identifying a face of an individual wearing the wearable item or segmenting a foreground silhouette of the wearable item from background image portions of the image, determining a portion of the wearable item identified within the image as being a patch portion representative of the wearable item depicted within the image, deriving one or more patterns of the wearable item based on image analysis of the determined patch portion of the image, deriving one or more colors of the wearable item based on image analysis of the determined patch portion of the image, and transmitting information regarding the derived one or more colors and information regarding the derived one or more patterns.

Abstract: Disclosed are a method and apparatus for training a classification model and a method and apparatus for classifying. A method for classifying comprises: extracting a feature from to-be-tested information inputted to a classification model having been trained; compressing the extracted feature into a low dimensional hidden feature capable of representing the to-be-tested information; performing decompression on the hidden feature to obtain a decompressed feature; performing rebuilding on the to-be-tested information based on the decompressed feature, to obtain reconstructed to-be-tested information; judging, based on a rebuild loss between the to-be-tested information and the reconstructed to-be-tested information, whether the to-be-tested information belongs to a known class or an unknown class; and performing classification on the to-be-tested information, via the classification model having been trained, in a case where it is determined that the to-be-tested information belongs to a known class.

Abstract: Methods, systems and programs for denoising a signal using discrete wavelet transformation are provided. For example, a method for denoising a signal may include determining a number of resolution levels to denoise, determining variable threshold(s) for each resolution level, applying the determined variable threshold(s) to denoise at least a detail component of each of the determined resolution levels. Each variable threshold includes a separately determined lower threshold and upper threshold. The method for denoising a signal may further include transforming, using an inverse discrete wavelet transformation, at least the denoised detail component for each of the determined resolution levels into a denoised signal.

Abstract: Information processing apparatus includes an input unit configured to input a captured image capturing a target object in a real environment, an obtaining unit configured to obtain an illumination condition in the real environment; a generation unit configured to generate a variation image as an image of a registration object reflecting the illumination condition based on three-dimensional data of the registration object and the obtained illumination condition, and a collation unit configured to collate the registration object with the target object based on a feature amount of the variation image and a feature amount of the captured image.

Abstract: Systems and methods for color selection are provided and include a mobile device having a mobile application configured to access a social media platform, retrieve a plurality of images from the social media platform, determine a dominant color for each image of the plurality of images, determine a closest matching paint color for the dominant color for each image, and display at least one of a color name and a color code associated with the closest matching paint color for the dominant color for each image.