Incorporating Graph-Based Models in a Deep Learning Framework for Operational Face Recognition

As submitted by the proposer: In classical face recognition, an input probe image is compared against a gallery of labeled face images in order to determine its identity. In most applications, the gallery images (identities) are assumed to be independent of each other, i.e., the relationship between gallery images is not exploited during the face recognition process. In this work, we propose a graph-based approach in which gallery images are used to generate a powerful network structure where the nodes correspond to individual identities (and consist of face images as well as biographic attributes such as gender, ethnicity, name, etc.) and the edge weights define the degree of similarity between two such nodes. This network can be used in several different ways: (a) to create clusters of identities based on graph clustering algorithms; (b) to predict the biographic and demographic attributes of an unknown probe image based on label propagation schemes; (c) to perform rapid recognition by restricting the search to only a fraction of the nodes in the graph; and (d) to infer missing information in nodes based on adjacent nodes that have strong edges. Experimental evaluation will be conducted on large datasets. ca/ncf

In classical face recognition, an input probe image is compared against a gallery of labeled face images in order to determine its identity. In most applications, the gallery images (identities) are assumed to be independent of each other, i.e., the relationship between gallery images is not exploited during the face recognition process.

In this work,the researcher proposes a graph-based approach in which gallery images are used to generate a powerful network structure where the nodes correspond to individual identities (and consist of face images as well as biographic attributes such as gender, ethnicity, name, etc.) and the edge weights define the degree of similarity between two such nodes. This network can be used in several different ways: (a) to create clusters of identities based on graph clustering algorithms; (b) to predict the biographic and demographic attributes of an unknown probe image based on label propagation schemes; (c) to perform rapid recognition by restricting the search to only a fraction of the nodes in the graph; and (d) to infer missing information in nodes based on adjacent nodes that have strong edges.

This project contains a research and/or development component, as defined in the applicable law.

nca/ncf