Kernel Face - toward unified theory of invariant feature extraction

摘要

In this paper, we propose a novel paradigm for invariant feature extraction in similar object recognition. The paradigm is "elimination of covariant element of feature makes invariant or robust feature against the variation". Effectiveness of the paradigm is partially confirmed against illumination variations. In this report, we apply the methodology deduced from the paradigm to image variation caused by pose change. From the experimental results, we could not obtain a rigid evidence for effectiveness of the methodology. The experimental results already showed potential of methodology. And the results clarify future problem.