Automatic face enhancement technique using sigmoid normalisation based on single scale Retinex algorithm

Abstract

The identical faces look dissimilar due to the change of illumination and it leads to incorrect results in face detection. In general, The Retinex algorithm improves visual quality by attaining good dynamic range and colour constancy. This work, a single-scale Retinex (SSR) algorithm based on sigmoid normalisation is proposed. The proposed normalisation suppresses the impact of outlier's data and improves the enhancement. The former min-max normalisation leads to loss of information and focused on specific parts of the normalised range. The SSR computes the reflectance values from the input image in different colour space. Followed, the sigmoid normalisation achieves good dynamic range compression through contrast factor (C) from estimated reflectance value. The performance of the proposed algorithm is validated in MUCT and IMM database. The experimental result shows the proposed algorithm attains better enhancement in contrast factor (C) between 40-50 scales under different illumination.