Wildfire simulation is implemented mainly by raster-based methods, with cellular automata (CA) being among the most widely used, due to their computational efficiency and easy implementation. However, most of the CA-based models often cause spatial distortions and artifacts, arising from the discrete nature of time and space in the simulation engine. Limited access to crucial environmental information in real time can further reduce their accuracy. In this work, we present an improved CA-based method which applies a perimeter shape correction mechanism during the simulation, so as to enhance wildfire simulation, by incorporating not only local transition rules, but also actions in the macroscopic level, leading to a more realistic representation of fire spread, The performance of the enhancement is validated using a wildfire test case scenario based on real-world wildfire data.