In retinal fundus images, the vertical cup-to-disc ratio (vCDR) is considered a crucial indicator for early screening and clinical diagnosis of glaucoma, and accurate segmentation of the optic disc (OD) and optic cup (OC) plays a vital role in its calculation. However, the segmentation performance of existing methods may be limited due to the complexity of clinical data and the challenge of distinguishing OC from OD, especially when OC boundary is blurred. To accurately calculate the vCDR, this paper proposes a novel diffusion model combined with wavelet transform for OD and OC segmentation. Specifically, we introduce a novel dual-branch denoising network that takes the retinal fundus image as prior information and parallelly integrates noise and semantic information within the network effectively. We also propose a cascaded discrete wavelet transform module to mitigate the negative impact of high-frequency noise components during the denoising process. Moreover, a new loss function is introduced that incorporates OC boundary information during the training process, emphasizing the boundary while gradually diminishing the significance of other regions. Finally, we validate the effectiveness of the proposed method for OD and OC segmentation using two publicly available datasets.