In an approach to integrating real-world properties into machine learning training, a real-world image is received. The real-world image is compared to a simulated image, where the comparison is performed using a discriminator network of a generative adversarial network (GAN). A generator network of the GAN is trained with results of the comparison of the real-world image to the simulated image. Responsive to determining that the real-world image is not optimal, the real-world image is iteratively tuned, using the generator network of the GAN, until it is determined that the real-world image is optimal, where the real-world image is optimal if the real-world image meets a predetermined threshold for accuracy of one or more image parameters of the simulated image versus the real-world image. The discriminator network of the GAN is trained with the real-world image.

An image processing method, for an identity document that comprises a data page, comprises acquiring a digital image of the page of data of the identity document. The method further comprises assigning a class or a super-class to the candidate identity document via automatic classification of the digital image by a machine-learning algorithm trained beforehand on a set of reference images in a training phase; processing the digital image to obtain a set of at least one intermediate image the weight of which is lower than or equal to the weight of the digital image; applying discrimination to the intermediate image using a discriminator neural network; and generating an output signal as output from the discriminator neural network, the value of which is representative of the probability that the candidate identity document is an authentic document or a fake.

A method for training an image classification model according to an embodiment includes training a feature extractor and a rotation angle classifier to predict a rotation angle of each of unlabeled first training images, training the image classification model to predict a label and rotation angle of each of labeled second training images, but predict a uniform label even though an actual rotation angle of each of the second training images is changed, generating a pseudo label based on a training image that satisfy a preset condition among unlabeled candidate images, and training the image classification model to predict a rotation angle of each of the third training images, and predict a label of each of the third training images based on the pseudo label, but predict a uniform label even though an actual rotation angle of each of the third training images is changed.

The pose and shape of a human body may be recovered based on joint location information associated with the human body. The joint location information may be derived based on an image of the human body or from an output of a human motion capture system. The recovery of the pose and shape of the human body may be performed by a computer-implemented artificial neural network (ANN) trained to perform the recovery task using training datasets that include paired joint location information and human model parameters. The training of the ANN may be conducted in accordance with multiple constraints designed to improve the accuracy of the recovery and by artificially manipulating the training data so that the ANN can learn to recover the pose and shape of the human body even with partially observed joint locations.

A computing system generates a training data set for training the prediction model to detect defects present in a target surface of a target specimen and training the prediction model to detect defects present in the target surface of the target specimen based on the training data set. The computing system generates the training data set by identifying a set of images for training the prediction model, the set of images comprising a first subset of images. A deep learning network generates a second subset of images for subsequent labelling based on the set of images comprising the first subset of images. The deep learning network generates a third subset of images for labelling based on the set of images comprising the first subset of images and the labeled second subset of images. The computing system continues the process until a threshold number of labeled images is generated.

An image processing method, for an identity document that comprises a data page, comprising comprises acquiring a digital image of the page of data of the identity document. The method further comprises assigning a class or a super-class to the candidate identity document via automatic classification of the digital image by a machine-learning algorithm trained beforehand on a set of reference images in a training phase; processing the digital image to obtain a set of at least one intermediate image the weight of which is lower than or equal to the weight of the digital image; applying discrimination to the intermediate image using a discriminator neural network; and generating an output signal as output from the discriminator neural network, the value of which is representative of the probability that the candidate identity document is an authentic document or a fake.

A discrimination device includes a sub-data set extraction unit for extracting from a plurality of labeled learning data a sub-learning data set to be used for learning and a sub-verification data set to be used for verification, a learning unit for performing supervised learning on the basis of the sub-learning data set to generate a pre-trained model for discriminating a label from data related to an object, a discrimination unit for conducting a discrimination processing using the pre-trained model on each piece of learning data contained in the sub-verification data set, a verification result recording unit for recording a result of the discrimination processing in association with the learning data, and a correctness detection unit for detecting learning data attached with a label that may be incorrect based on the discrimination processing results recorded in association with respective learning data.

A control unit is disclosed to control an imaging unit to perform imaging of a tray/container. The control unit can cause the performance of actions on the container using automated machines and/or directing humans to perform an action. For example, the presence of contamination in a container can be detected based on an image of the container captured by an imaging unit. The control unit can receive an image of the container from the imaging unit, determine whether the container is contaminated, and direct the container to a cleaning unit. Moreover, the control unit can detect a product based on an image of the product, determine an identity of the product based on the received image, and command an indicating unit to indicate a failure to determine the identity of the product.

The disclosure discloses a method and a system for generating image adversarial examples based on an acoustic wave. The method includes: acquiring an image containing a target object or a target scene; generating simulated image examples for the acquired image, wherein the simulated image examples have adversarial effects on a deep learning algorithm in a target machine vision system; optimizing the generated simulated image examples to obtain an optimal adversarial example and corresponding adversarial parameters; and injecting the adversarial parameters into an inertial sensor of the target machine vision system in a manner of an acoustic wave, such that the adversarial parameters are used as sensor readings that will cause an image stabilization module in the target machine vision system to operate to generate particular blurry patterns in a generated real-world image so as to generate an image adversarial example in a physical world.

Computer vision systems and methods for localizing image forgery are provided. The system generates a constrained convolution via a plurality of learned rich filters. The system trains a convolutional neural network with the constrained convolution and a plurality of images of a dataset to learn a low level representation of each image among the plurality of images. The low level representation is indicative of a statistical signature of at least one source camera model of each image. The system can determine a splicing manipulation localization by the trained convolutional neural network.