Robust Features for Frontal Face Authentication in Difficult Image Conditions

In this report we extend the recently proposed {DCT}-mod2 feature extraction technique (which utilizes polynomial coefficients derived from {2D} {DCT} coefficients obtained from horizontally & vertically neighbouring blocks) via the use of various windows and diagonally neighbouring blocks. We also evaluate enhanced {PCA}, where traditional {PCA} feature extraction is combined with {DCT}-mod2. Results using test images corrupted by a linear and a non-linear illumination change, white Gaussian noise and compression artefacts, show that use of diagonally neighbouring blocks and windowing is detrimental to robustness against illumination changes while being useful for increasing robustness against white noise and compression artefacts. We also show that the enhanced {PCA} technique retains all the positive aspects of traditional {PCA} (that is robustness against white noise and compression artefacts) while also being robust to illumination direction changes; moreover, enhanced {PCA} outperforms {PCA} with histogram equalisation pre-processing.