basically you are correction for an inflation of your test statistic (here in Genome wide association studies). If you assume that most of your SNPs are NOT assciated with the trait of interest, then the overwhelming majority of statistical tests'p-values (one per SNP) should follow a uniform distribution (H0). Similarly, the vast majority of your tests (this is mainly in this context that we use GC) should follow of chi-square with a mean of 1. Even in the presence of, say 100 true positives, the mean of all your tests should not be very far from 1. If they are, this can be due to population stratification, diffrential DNA quality (in case controls), or even to polygenic component (infinity of very low effects).
Whatever the reason, this can give an excess of false positive results and on way to correct is to create a new statistic based on the mean of your chi-square distribution (or the median or other robust moments). The simplest way is to divide all your test statistics by the mean of the test statistics.
for SNP i, if your statistic is 2 and and the mean of all your stats is 1.3 (typical of an inflation in case-control studies), then your new corrected statistics will be
2/1.3 = 1.58. This is one of all the possible corrections.
Your new statistic is weaker than the original one (usually if the mean is below 1, we are not correcting - ie not creating a higher chi-square).
Hope this helps