Face Verification With Balanced Thresholds

Abstract

Abstract-The process of face verification is guided by a prelearned global threshold, which, however, is often inconsistent with class-specific optimal thresholds. It is, hence, beneficial to pursue a balance of the class-specific thresholds in the model-learning stage. In this paper, we present a new dimensionality reduction algorithm tailored to the verification task that ensures threshold balance. This is achieved by the following aspects. First, feasibility is guaranteed by employing an affine transformation matrix, instead of the conventional projection matrix, for dimensionality reduction, and, hence, we call the proposed algorithm threshold balanced transformation (TBT). Then, the affine transformation matrix, constrained as the product of an orthogonal matrix and a diagonal matrix, is optimized to improve the threshold balance and classification capability in an iterative manner. Unlike most algorithms for face verification which are directly transplanted from face identification literature, TBT is specifically designed for face verification and clarifies the intrinsic distinction between these two tasks. Experiments on three benchmark face databases demonstrate that TBT significantly outperforms the state-of-the-art subspace techniques for face verification.

Person recognition for service robotics applications

“…Many different projections have been proposed including famous examples like Eigenfaces [4], where the basis of the face space is spanned by eigenvectors obtained from a PCA analysis of the training images, and Fisherfaces [7], where the projection bases on Fisher's Linear Discriminant and is computed to be nearly orthogonal to intra-class scatter to minimize intra-class variance while maximizing interclass variance. Further projection methods construct linear subspaces for each person based on the original gallery images and measure the distance of query faces to these spaces [12], enhance the projection by kernel methods to model higher-order pixel dependencies [14], or refine the Fisherfaces approach with more intricate target subspaces [13], [15]. All projection methods are characterized by a low computational complexity and all modern variants and Fisherfaces attain high recognition rates on standard databases.…”

Facial feature localization based on an improved active shape model

“…The results of the localization can be used for the human-machine interface [2,13,14,21], the recognition of a face and facial expression [6,15,17,23,24], 3D face recognition [1], and etc. Facial feature location methods can be classified into the following two categories: (1) local methods [7,16], which detect eye, mouth, chin separately; (2) global methods [3,4,22], which detect facial features jointly.…”

Visual recognition by subspace approaches on LBP features

“…Subspace face recognition can be formulated as global formulation or class-specific formulation. In global formulation, one unified projection matrix is trained for all the classes through constructing a pooled covariance matrix [116]. In class-specific formulation, a classspecific projection matrix is trained for each class [117].…”