A method for generating object and part detectors includes accessing a collection of training images. The collection of training images includes images annotated with an object label and images annotated with a respective part label for each of a plurality of parts of the object. Joint appearance-geometric embeddings for regions of a set of the training images are generated. At least one detector for the object and its parts is learnt using annotations of the training images and respective joint appearance-geometric embeddings, e.g., using multi-instance learning for generating parameters of scoring functions which are used to identify high scoring regions for learning the object and its parts. The detectors may be output or used to label regions of a new image with object and part labels.

A biometric analysis structure, method and neural network with coded mask are provided. The biometric analysis structure includes a display panel, a light source, and a sensor. The sensor is disposed on the optical path of light from the light source and reflected by the display panel. The biometric analysis structure includes an coded mask. The coded mask is disposed on the optical path in front of the sensor. the coded mask is represented as a first matrix in a matrix form and the first matrix is a delta function after satisfying an autocorrelation operation. The resulting image can be inversely resolved based on the image of the coded mask. Thus, the security of the fingerprint recognition method is improved, and the thickness of the entire imaging structure is reduced.

An object recognition method is provided. The method includes: detecting an occlusion region of an object in an image, to obtain a binary image; obtaining occlusion binary image blocks; querying a mapping relationship between occlusion binary image blocks and binary masks included in a binary mask dictionary to obtain binary masks corresponding to the occlusion binary image blocks; synthesizing the binary masks queried based on each of the occlusion binary image blocks, to obtain a binary mask corresponding to the binary image; and determining a matching relationship between the image and a prestored object image, based on the binary mask corresponding to the binary image, a feature of the prestored object image, and a feature of the to-be-recognized image.

Systems, methods, and non-transitory media are provided for providing spatiotemporal recycling networks (e.g., for video segmentation). For example, a method can include obtaining video data including a current frame and one or more reference frames. The method can include determining, based on a comparison of the current frame and the one or more reference frames, a difference between the current frame and the one or more reference frames. Based on the difference being below a threshold, the method can include performing semantic segmentation of the current frame using a first neural network. The semantic segmentation can be performed based on higher-spatial resolution features extracted from the current frame by the first neural network and lower-resolution features extracted from the one or more reference frames by a second neural network. The first neural network has a smaller structure and/or a lower processing cost than the second neural network.

Automatic nuclear segmentation. In an embodiment, a plurality of superpixels are determined in a digital image. For each of the superpixels, any superpixels located within a search radius from the superpixel are identified, and, for each unique local combination between the superpixel and any identified superpixels located within the search radius from the superpixel, a local score for the local combination is determined. One of a plurality of global sets of local combinations with an optimum global score is identified based on the determined local scores.

A medical information processing apparatus according to an embodiment includes a storage and processing circuitry. The storage is configured to store therein, for each point of a frequency space represented by a plurality of pieces of first frequency component data acquired by applying frequency conversion to data inside regions of interest set to medical images, characteristic data representing a tendency of spectral values that appear at the point. The processing circuitry is configured to acquire second frequency component data by applying frequency conversion to a medical image to be processed, to determine similarity of a spectral value at each point of a frequency space represented by the second frequency component data, with the characteristic data, and to designate a target area in the frequency space represented by the second frequency component data based on the result of the determination.

According to one embodiment, an article recognition apparatus includes an image acquisition unit, a recognition unit, a region detection unit, a storage unit, and a determination unit. The recognition unit recognizes each of the articles. The region detection unit determines article region information. The storage unit stores article information including a reference value for the article region information. The determination unit determines that an unrecognized article exists, if the reference value for the article region information of each article which the recognition unit recognized does not match with the article region information.

A computing device may obtain a first captured image of a scene and a second captured image of the scene. For a plurality of m×n pixel tiles of the first captured image, the computing device may determine respective distance matrixes. The distance matrixes may represent respective fit confidences between the m×n pixel tiles and pluralities of target p×q pixel tiles in the second captured image. The computing device may approximate the distance matrixes with respective bivariate quadratic surfaces. The computing device may upsample the bivariate quadratic surfaces to obtain respective offsets for pixels in the plurality of m×n pixel tiles. The respective offsets, when applied to pixels in the plurality of m×n pixel tiles, may cause parts of the first captured image to estimate locations in the second captured image.

A method, computer system, and a computer program product for analyzing visual defects is provided. The present invention may include generating a template image. The present invention may include capturing a test image. The present invention may include performing an image registration between the template image and the test image. The present invention may include generating a registered test image. The present invention may include performing an image difference analysis between the registered test image and the template image. The present invention may include generating a differential image. The present invention may include synthesizing the registered, differential image, and template image. The present invention may include generating a synthetic image. The present invention may include inputting the synthetic image into a multi-scale detection network. The present invention may include generating a defect map.

A method of automatic defect classification (ADC) includes detecting defective parts from a substrate wherein at least one unit process is performed; and classifying defect types of the respective defective parts, wherein the classifying includes obtaining a scanning electron microscope (SEM) image of each of the defective parts; registering information about the substrate in a graphic data system (GDS) image corresponding to each SEM image; defining a plurality of defects of interest (DOIs) categorizing defects of the respective defective parts; defining a DOI rule that is a criterion for determining which defects of the respective defective parts correspond to which DOI from among the DOIs; and analyzing the image to classify which defects of the respective defective parts correspond to which DOI from among the DOIs according to the DOI rule.