Enrichment normalisation is useful in studies where you have a difference in enrichment efficiency between different samples. It is related to the percentile normalisation, where you specify a single reference percentile on which to normalise your data and the data is corrected by a single factor so that all samples have the same value at this reference point.
In enrichment normalisation you specify two normalisation percentiles. The lower percentile should represent the baseline value, and should be the last part of your distribution you consider to be unenriched. The second value should be the percentile at which you believe you have pretty much saturated your enrichment.
The way the enrichment normalisation works is that it applies a simple addition correction to get the different datasets to match exactly at the lower percentile. It then applies a multiplication correction so that the samples match at the upper percentile. The data is therefore scaled between the two defined reference points, which should correct for any differences in enrichment efficiency between samples.
Enrichment normalisation would most commonly be used on ChIP-Seq samples where the efficiency of the ChIP may vary between different experiments. It could also be used on other experiment types to try to even out other biases. We have used it for example on looking at genomic coverage in PBAT bisulphite libraries where the data has an overall GC bias which is somewhat different in different samples.