Semantic segmentation of images is an important task in computer vision, as it provides precise pixel labels that benefit various applications such as autonomous driving, medical imaging, and satellite image analysis by providing crucial information about the location and boundaries of objects. Segmenting a pixel requires a deep understanding of the surrounding scene, and the contextual information from nearby and distant pixels plays a crucial role. In this study, we conducted studies to show the impact of contextual information on the performance of segmentation models. Specifically, a method to preprocess the context and refine the input of the segmentation model is proposed, providing it with useful clues to accurately detect the pixels' labels. Additionally, a method for contextual post-processing is developed to improve the consistency of the model predictions based on object boundaries. Finally, a novel context-aware decoder called SPADe is provided, inspired by the human vision system, which effectively captures local and global contextual information with low computational cost. We demonstrate that our proposed solution leveraging contextual information significantly improves the performance of segmentation models compared to the state-of-the-art models.