Carnegie Mellon University

Dipan Kumar Pal

Marios Savvides

Yutong Zheng

Felix Juefei-Xu

Raied Aljadaany

A method of training a deep neural network, such as would be used in facial recognition, includes training the deep neural network to normalize feature vectors to a learned value representing the radius of a multi-dimensional hypersphere using a convex augmentation of the primary loss function.

Convex feature normalization for face recognition

During training of deep neural networks, a Copernican loss (L) is designed to augment a primary loss function, for example, a standard Softmax loss, to explicitly minimize intra-class variation and simultaneously maximize inter-class variation. Copernican loss operates using the cosine distance and thereby affects angles leading to a cosine embedding, which removes the disconnect between training and testing.

Discriminative cosine embedding in machine learning

Disclosed herein is a method for performing blind deconvolutions of blurred images. The method approximates the proximal operators for the data fidelity term and the prior term of a minimization function using trained neural networks and solves the minimization using iterations of the Douglas-Rachford algorithm.

Method and architecture for blind image deconvolution

During training of deep neural networks, a Copernican loss (L) is designed to augment the standard Softmax loss to explicitly minimize intra-class variation and simultaneously maximize inter-class variation. Copernican loss operates using the cosine distance and thereby affects angles leading to a cosine embedding, which removes the disconnect between training and testing.

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.

Growing community of inventors

Pittsburgh, PA, United States of America

Dipan Kumar Pal