Research Area 5: Task-Driven Coupled Dictionary Design for Heterogeneous Sensory Data

Abstract

The objective of this proposal is to create new methods of coupling multimodal data via sparsity for object classification. Unlike existing research on dictionary design based on single modality, this effort will design coupled discriminative dictionaries using unstructured multimodal data. Effort will create a pair of discriminative coupled dictionaries to capture latent shared information that exist across two different modalities. The latent information will be exploited to directly compare features by computing the sparse codes for the two modalities (i.e., thermal and visible face images of the same subject). A task-driven dictionary learning method will be designed to provide supervised discriminative dictionaries using the recently developed bi-level optimization techniques. Exploitation of nonlinear relationships between sensory data via kernel theory will be embedded in these algorithms.

Document Details

Document Type
DoD Grant Award
Publication Date
Jan 12, 2017
Source ID
W911NF1610083

Entities

People

  • Nasser M. Nasrabadi

Organizations

  • Army Contracting Command
  • United States Army
  • West Virginia University

Tags

Fields of Study

  • Computer science

Readers

  • Computational Linguistics
  • Computer Vision.
  • Distributed Systems and Data Platform Development