A tentative local score between a point in a feature image in a template image and a point, in a target object image, at a position corresponding to the point in the feature image is calculated, and a determination is performed as to whether the tentative local score is smaller than 0. In a case where the tentative local score is greater than or equal to 0, the tentative local score is employed as a local score. In a case where the tentative local score is smaller than 0, the tentative local score is multiplied by a coefficient and the result is employed as a degree of local similarity.

A method for detecting an object includes inputting information of a moving object included in a plurality of images and generating a regression tree. In response to input of a new image, the system communicates information of a moving object included in the newly inputted image into the regression tree, and determines a size of a person included in the new image based on a resultant value of the regression tree.

A method, apparatus, non-transitory computer readable medium, and system for panoptic segmentation are described. Embodiments may generate a feature pyramid for an input image, wherein the feature pyramid comprises a plurality of feature maps at different resolution levels, apply an attention module to the feature pyramid to produce an enhanced feature map, combine the enhanced feature map with each of the plurality of feature maps to produce an enhanced feature pyramid, generate semantic information for the input image based on the enhanced feature pyramid, generate a plurality of object regions based on the enhanced feature pyramid, generate instance information for each of the plurality of object regions, and generate panoptic segmentation information for the input image based on the semantic information and the instance information for each of the plurality of object regions.

To improve a search accuracy in template matching based on a coarse-to-fine search method, a first pyramid is generated by hierarchizing a plurality of template images having different resolutions from each other in order of the resolution, a second pyramid is generated by hierarchizing a plurality of search target images having different resolutions from each other corresponding to the same resolutions as the template images in the first pyramid in order of the resolution, and a search process of searching for a feature image included in the template image of the first pyramid from the search target image in the second pyramid in a hierarchy of the same resolution. When the first pyramid is generated, the number of stages of the first pyramid is set such that a size of the feature image included in each of the template images is not smaller than a previously set threshold.

An information processing device according to the present invention includes: a proper identifier output unit which outputs proper identifiers for identifying learning images; a feature vector calculation unit which calculates feature vectors of at least a part of patches included in registered patches that are registered in a dictionary for compositing a restored image; and a search similarity calculation unit which calculates a similarity calculation method that classifies the proper identifiers to be given to the registered patches based on the feature vectors.

A computer-implemented method and system register plural images using a computer processor and computer memory, where computer code stored in the computer memory causes the computer processor to perform the registration. The registration includes receiving a reference image; receiving a sensed image; computing an approximate image transformation between the reference image and the sensed image; finding plural candidate tie points by utilizing template matching; applying radiometric filtering to at least a portion of the plural candidate tie points; applying geometric filtering to at least a portion of the plural candidate tie points; computing a calculated image transformation using candidate points of the plural candidate points that were not filtered out by the radiometric and geometric filtering; transforming the sensed image using the calculated image transformation; and registering the sensed image with the reference image.

Systems and methods are provided for embedded data inference. The systems and methods may process camera and other sensor data in by leveraging processing and storage capacity of one or more devices nearby or in the cloud to augment or update the sensor processing of an embedded device. The joint processing may be used in stationary cameras or in vehicular systems such as cars and drones, and may improve crop assessments, navigation, and safety.

An image is passed through an image identifier to identify a coarse category for the image and a bounding box for a categorized object. A mask is used to identify the portion of the image that represents the object. Given the foreground mask, the convex hull of the mask is located and an aligned rectangle of minimum area that encloses the hull is fitted. The aligned bounding box is rotated and scaled, so that the foreground object is roughly moved to a standard orientation and size (referred to as calibrated). The calibrated image is used as an input to a fine-grained categorization module, which determines the fine category within the coarse category for the input image.

There is provided a framework including systems and methods for analyzing skin parameters from images or videos showing skin. Using a series of Hierarchical Differential Image Filters (HDIF), it becomes possible to detect different skin features such as wrinkles, spots, and roughness. The hierarchical differential image filter computes two enhancements to an image showing skin at two different levels of enhancement, determines a differential image using the two enhancements and computes the skin analysis rating using the differential image. These skin ratings are comparably accurate to actual ratings by dermatologists.

A method and apparatus for vascular disease detection and characterization using a recurrent neural network (RNN) is disclosed. A plurality of 2D cross-section image patches are extracted from a 3D computed tomography angiography (CTA) image, each extracted at a respective sampling point along a vessel centerline of a vessel of interest in the 3D CTA image. Vascular abnormalities in the vessel of interest are detected and characterized by classifying each of the sampling points along the vessel centerline based on the plurality of 2D cross-section image patches using a trained RNN.