A method of detecting concealed objects using a thermal imager includes obtaining an output comprising a plurality of pixels representing a person, analyzing each pixel matching a contour of the person and excluding any pixel within a blob bounding box of the person, and determining whether a pixel address is represented in a pixel map. In addition, the method includes comparing a value of each remaining pixel to an allowable minimum threshold value representing a lower pre-defined body temperature, and comparing the value of each remaining pixel greater than or equal to the allowable minimum threshold value to an upper allowable threshold value representing an upper pre-defined body temperature. The method also includes excluding any of the remaining pixels within a range between the lower and upper pre-defined body temperatures to define final set of pixels and calculating a pixel difference to indicate a severity of the difference.

Apparatus for diagnosing breast cancer, the apparatus comprising a controller having a set of instructions executable to: acquire a contrast enhanced region of interest (CE-ROI) in an X-ray image of a patient's breast, the X-ray image comprising X-ray pixels that indicate intensity of X-rays that passed through the breast to generate the image; determine a texture neighborhood for each of a plurality of X-ray pixels in the CE-ROI, the texture neighborhood for a given X-ray pixel of the plurality of X-ray pixels extending to a bounding pixel radius of BPR pixels from the given pixel; generate a texture feature vector (TF) having components based on the indications of intensity provided by a plurality of X-ray pixels in the CE-ROI that are located within the texture neighborhood; and use a classifier to classify the texture feature vector TF to determine whether the CE-ROI is malignant

Methods and systems for training a metric learning convolutional neural network (CNN)-based model for cross-domain image retrieval are disclosed. The methods and systems perform steps of generating a plurality of batches sampled from a cross-domain training dataset to train the CNN-based model to match images of different sub-categories from one domain to another, and training the CNN-based model using the generated batches. The CNN-based model comprises various pooling, normalization, and concatenation layers that enable it to concatenate the normalized outputs of multiple concatenation layers. Use of the generated batches comprises executing a loss function based on one or more batches, where the loss function is a triplet, contrastive, or cluster loss function. Embodiments of the present invention enable the CNN-based model to summarize information from multiple convolutional layers, thus improving visual search. Also disclosed are benefits of the new methods, and alternative embodiments of implementation.

An annotation device, comprising: a display that performs sequential playback display of a plurality of images that may contain physical objects that are the subject of annotation, and a processor that acquires specific portions that have been designated within the images displayed on the display as annotation information, sets operation time or data amount for designating the specific portions, and at a point in time where designation of the specific portions has been completed for the operation time, a time based on data amount, or data amount, that have been set, requests learning to an inference engine that creates an inference model by learning, using annotation information that has been acquired up to the time of completion as training data representing a relationship between the physical object and the specific portions.

Systems and methods are provided for end-to-end learning of commands for controlling an autonomous vehicle. A pre-processor pre-processes image data acquired by sensors at a current time step (CTS) to generate pre-processed image data that is concatenated with additional input(s) (e.g., a segmentation map and/or optical flow map) to generate a dynamic scene output. A convolutional neural network (CNN) processes the dynamic scene output to generate a feature map that includes extracted spatial features that are concatenated with vehicle kinematics to generate a spatial context feature vector. An LSTM network processes, during the (CTS), the spatial context feature vector at the (CTS) and one or more previous LSTM outputs at corresponding previous time steps to generate an encoded temporal context vector at the (CTS). The fully connected layer processes the encoded temporal context vector to learn control command(s) (e.g., steering angle, acceleration rate and/or a brake rate control commands).

An image generation method and a computing device using the method, includes creating an image database with a plurality of original images, and obtaining a plurality of first outline images of an object by detecting an outline of the object in each of the original images. Numerous first feature matrixes are obtained by calculating a feature matrix of each of the first outline images. A second feature matrix of a second outline image input by a user is calculated. A target feature matrix is selected from the plurality of first feature matrixes, the target feature matrix has a minimum difference as the second feature matrix. A target image corresponding to the target feature matrix is matched and displayed from the image database. The method and device allow detection of an object outline in an image input by users and the generation of an image with the detected outline.

Improvements are disclosed for detecting counterfeit objects, based on comparison to digital fingerprints that describe features found in images of objects known to be counterfeit.

The invention relates to an apparatus for identifying a candidate object in image data and determining a likelihood that the candidate object is an object from an object class. The apparatus comprises an image data receiving unit for receiving image data of an object of the object class, a seed element selecting unit for selecting a portion of the image elements as seed elements, a contour point identifying unit for identifying, for each seed element (SE), contour points, the contour points of a seed element circumscribing a candidate object which comprises the seed element, and a seed score determining unit for determining, for each seed element, a seed score indicative of a likelihood that the candidate object is an object from the object class. The invention allows differentiation between an object of an object class of interest and artifacts.

A method includes receiving an image from a mammogram, removing noise from the image, computing a point of interest on the de-noised image, creating a mesoscale region of interest on the de-noised image, computing a connectivity for the mesoscale region of interest, identifying a connected component using the computed connectivity, where the connected component represents a branch of a global curvilinear structure, selecting a set of branches based on a physical property for each branch of the global curvilinear structure, pruning each branch based on an error-tolerant, adaptive polynomial fit, identifying remaining regions of interest in each pruned branch, and growing a chain formed by remaining points of interest included in the remaining regions of interest, where the chain represents a macroscopic, global curvilinear calcified arterial structure. The quantitation of the calcified arterial structures may be used as a biomarker for risk stratification of heart disease.

An electronic apparatus includes a processor configured to obtain a plurality of images, extract deep features with respect to the plurality of images using a feature extraction model, classify the plurality of images into certain groups using the extracted deep features and a classification model, display a result of the classification on the display, determine whether the feature extraction model and/or the classification model need to be updated using the result of the classification, and train and update at least one of the feature extraction model and the classification model based on a result of the determination. The electronic apparatus may estimate a deep feature of an image using a rule-based or artificial intelligence (AI) algorithm. When the deep feature of the image is estimated using the AI algorithm, the electronic apparatus may use a machine learning, neural network, or deep learning algorithm, or the like.