The binomial filter is related to the ChiSquare filter in that it doesn't use the current quantitation of your data, but instead is based around the proportion of forward and reverse reads within your probes. It is most commonly used for the analysis of bisulphite sequencing data but could be used for any type of data where a change in strand proportion is the major interest.
The difference between this test and the chi-square test is that in the chi-square filter the expectation is that there is no global change between your two samples. The binomial filter can be used in samples where there is a global change and you are interested in finding probes whose for/rev ratio changes in a way which is unusual compared to what is happening globally.
The test works in two stages. You select two data stores, a from store and a to store. In the first stage the filter creates a set of mean for/rev proportion values where it finds the mean for/rev proportion in your 'to' store from all starting proportions in your 'from' stores, so it might for example find that where you started with 60% forward in your from store you ended up with (on average) 80% forward in your 'to' store for the same probes.
In the second part of the test the filter does a second pass through your data testing the actual observed for/rev proportion to the expected value from the first part using a binomial test. For example if you had an individual probe which had 60% forward in your 'from' store it might have 90% forward in your 'to' store from 200 observations, so a binomial test can be used to determine whether this is significantly different to what's happening globally.