QRCP-based preprocessing for illumination invariant measure under severe illumination variations

The previous illumination invariant measure performs unsatisfactorily under severe illumination variations, since severe illumination variations cause large differences of illumination intensities in the face local region, especially for the edge of cast shadows. In this paper, we conduct orthogonal triangular with column pivoting (QRCP) decomposition in the N xN local block which generates N QRCP coefficients. QRCP unit is defined as the quotient of the largest QRCP coefficient and the sum of all QRCP coefficients in the local block. To further control the illumination effect, we obtain corrected QRCP unit by calculating the gamma power of QRCP coefficients. We define QRCP Image as the combination of corrected QRCP units. Finally, QRCP Image is employed as a preprocessing method of illumination invariant measures (LNN-face/BGIR-face/EGIR-face). The experimental results on the illumination-varying face databases (Extended Yale B and CMU PIE) indicate that the proposed method is efficient to tackle severe illumination variations.

Automated summary

This paper proposes an illumination invariant measure based on orthogonal triangular with column pivoting (QRCP) decomposition. By generating QRCP coefficients and further correction, it can effectively handle severe illumination variations.