Single-source domain generalization (SDG) in medical segmentation is a challenging yet essential task due to the ubiquitous domain shifts across clinical image datasets. In this paper, we present a novel illumination enhancement-based domain generalization framework to improve the generalization capability of the model on unseen test datasets. Specifically, we first develop an image decomposition network (IDN) to decompose polyp images into reflectance, local illumination, and global illumination components, respectively. To derive images with various global illuminations, we further propose an illumination transform network (ITN) to generate target-like global illumination maps. We extensively evaluate the proposed model in polyp segmentation performance on four colonoscopy datasets: CVC-ClinicDB, CVC-ColonDB, ETIS-Larib, and Kvasir-SEG. Our proposed method obtains satisfying segmentation results on unseen datasets and outperforms other image enhancement and domain generalization methods.