In the Tanner graph of low density parity check (LDPC) code, there is a kind of structure called trapping set whose uncorrected variable nodes may not be corrected through iteration. In this paper, we will introduce a novel coding and decoding scheme of generalized LDPC codes with the assistance of quadratic residue code. In this scheme, by using quadratic residue code and sacrificing a bit of rate, some variable nodes will be selected to be given the additional protection, hoping that this scheme would help weaken the effects of trapping sets and improve the performance of the original LDPC codes. In the simulation, after counting uncorrected bits and analyzing the related error location histogram, we find that some bits or variable nodes may incline to be wrong. For this result, we figure out its reason from the perspective of trapping set, and select these variable nodes to construct the generalized LDPC codes. From our simulation, this scheme mitigates the error floor of some LDPC codes effectively.