Abstract: The invention proposes a method of training a convolutional neural network in which, at each convolution layer, weights for one seed convolutional filter per layer are updated during each training iteration. All other convolutional filters are polynomial transformations of the seed filter, or, alternatively, all response maps are polynomial transformations of the response map generated by the seed filter.

Abstract: The invention specifies a method of improving a subsequent iterations of a generative network by adding a ranking loss to the total loss for the network, the ranking loss representing the marginalized difference between a discriminator score for a generated image in one iteration of the generative network and the discriminator score for a real image from a subsequent iteration of the generative network.

Abstract: The invention proposes a method of training a convolutional neural network in which, at each convolution layer, weights for one seed convolutional filter per layer are updated during each training iteration. All other convolutional filters are polynomial transformations of the seed filter, or, alternatively, all response maps are polynomial transformations of the response map generated by the seed filter.

Abstract: The invention specifies a method of improving a subsequent iterations of a generative network by adding a ranking loss to the total loss for the network, the ranking loss representing the marginalized difference between a discriminator score for a generated image in one iteration of the generative network and the discriminator score for a real image from a subsequent iteration of the generative network.

Abstract: The invention proposes a method of training a convolutional neural network in which, at each convolutional layer, weights for one seed convolutional filter per layer are updated during each training iteration. All other convolutional filters are polynomial transformations of the seed filter, or, alternatively, all response maps are polynomial transformations of the response map generated by the seed filter.

Abstract: Perturbative neural networks are comprised of one or more modules, called perturbation layers which can be used as an alternative to a convolutional layer. The perturbation layer does away with convolution in the traditional sense and instead computes its response as a weighted linear combination of non-linearly activated additive noise perturbed inputs.

Abstract: Identifying a masked suspect is one of the toughest challenges in biometrics that exist. This is an important problem faced in many law-enforcement applications on almost a daily basis. In such situations, investigators often only have access to the periocular region of a suspect's face and, unfortunately, conventional commercial matchers are unable to process these images in such a way that the suspect can be identified. Herein, a practical method to hallucinate a full frontal face given only a periocular region of a face is presented. This approach reconstructs the entire frontal face based on an image of an individual's periocular region. By using an approach based on a modified sparsifying dictionary learning algorithm, faces can be effectively reconstructed more accurately than with conventional methods. Further, various methods presented herein are open set, and thus can reconstruct faces even if the algorithms are not specifically trained using those faces.

Abstract: Identifying a masked suspect is one of the toughest challenges in biometrics that exist. This is an important problem faced in many law-enforcement applications on almost a daily basis. In such situations, investigators often only have access to the periocular region of a suspect's face and, unfortunately, conventional commercial matchers are unable to process these images in such a way that the suspect can be identified. Herein, a practical method to hallucinate a full frontal face given only a periocular region of a face is presented. This approach reconstructs the entire frontal face based on an image of an individual's periocular region. By using an approach based on a modified sparsifying dictionary learning algorithm, faces can be effectively reconstructed more accurately than with conventional methods. Further, various methods presented herein are open set, and thus can reconstruct faces even if the algorithms are not specifically trained using those faces.

Abstract: Determining a match between the subjects of first and second images as a function of decimal-number representations of regions of the first and second images. The decimal-number representations are generated by performing discrete transforms on the regions so as to obtain discrete-transform coefficients, performing local-bit-pattern encoding of the coefficients to create data streams, and converting the data streams to decimal numbers. In one embodiment, the first and second images depict periocular facial regions, and the disclosed techniques can be used for face recognition, even where a small portion of a person's face is captured in an image. Subspace modeling may be used to improve accuracy.

Abstract: Determining a match between the subjects of first and second images as a function of decimal-number representations of regions of the first and second images. The decimal-number representations are generated by performing discrete transforms on the regions so as to obtain discrete-transform coefficients, performing local-bit-pattern encoding of the coefficients to create data streams, and converting the data streams to decimal numbers. In one embodiment, the first and second images depict periocular facial regions, and the disclosed techniques can be used for face recognition, even where a small portion of a person's face is captured in an image. Subspace modeling may be used to improve accuracy.