Outlier-robust feature scaling

Dear All,

What feature scaling method would you recommend to bring data within [0 1] or [-1 1], for features that are heavily skewed /non nonnormal (t location scale, generalized extreme value distros), and for which outliers are preferably not removed (features are i1-i2 image subtractions that use outliers in classification, some also have different units).

Thank you,

OL