Diffusion-Based Purification (DBP) has emerged as an effective defense mechanism against adversarial attacks. The success of DBP is often attributed to the forward diffusion process, which reduces the distribution gap between clean and adversarial images by adding Gaussian noise. Although this explanation is theoretically grounded, the precise contribution of this process to robustness remains unclear. In this paper, through a systematic investigation, we propose that the intrinsic stochasticity in the DBP procedure is the primary factor driving robustness. To explore this hypothesis, we introduce a novel Deterministic White-Box (DW-box) evaluation protocol to assess robustness in the absence of stochasticity, and analyze attack trajectories and loss landscapes. Our results suggest that DBP models primarily leverage stochasticity to evade effective attack directions, and that their ability to purify adversarial perturbations can be weak. To further enhance the robustness of DBP models, we propose Adversarial Denoising Diffusion Training (ADDT), which incorporates classifier-guided adversarial perturbations into diffusion training, thereby strengthening the models' ability to purify adversarial perturbations. Additionally, we propose Rank-Based Gaussian Mapping (RBGM) to improve the compatibility of perturbations with diffusion models. Experimental results validate the effectiveness of ADDT. In conclusion, our study suggests that future research on DBP can benefit from the perspective of decoupling stochasticity-based and purification-based robustness.

Superpixels provide an efficient representation of images, and are applicable for subsequent vision tasks. In this paper, we propose an edge-aware superpixel algorithm based on an unsupervised convolutional neural network (CNN). Noticing that to adhere the boundaries of objects is one of the most essential characteristics of superpixels, we propose an entropy-based edge-aware term, which helps fit the differential model of the pixel-superpixel soft-assignment matrix predicted from CNN to image gradients, i.e. generate boundary-aligning superpixels. The proposed algorithm yields more boundary-adhering superpixels, and experimental results on BSDS500 show the effectiveness of the proposed edge-aware term.