We propose a semi-blind decision-directed joint channel estimation and decoding algorithm for low-density parity check (LDPC) coded Orthogonal Frequency Division Multiplexing (OFDM) systems. There are two novel aspects in our algorithm - (i) an enhanced semi-blind channel estimation algorithm that hypothesizes a small set of symbols as additional pilots and (ii) an edge strength-based metric to determine a small subset of decisions at the output of the LDPC decoder as being reliable. Our algorithm has an order of magnitude lower complexity than a recently proposed machine learning-based channel estimation algorithm, while their performances are comparable.