Abstract
We propose an explicitly discriminative and ‘simple’ ap-proach to generate invariance to nuisance transformationsmodeled as unitary. In practice, the approach works well tohandle non-unitary transformations as well. Our theoreti-cal results extend the reach of a recent theory of invarianceto discriminative and kernelized features based on unitarykernels. As a special case, a single common frameworkcan be used to generate subject-specific pose-invariant fea-tures for face recognition and vice-versa for pose estima-tion. We show that our main proposed method (DIKF)can perform well under very challenging large-scale semi-synthetic face matching and pose estimation protocols with unaligned faces using no landmarking whatsoever. We additionally benchmark on CMU MPIE and outperform previous work in almost all cases on off-angle face matching while we are on par with the previous state-of-the-art on theLFW unsupervised and image-restricted protocols, withoutany low-level image descriptors other than raw-pixels.