<p>Multiple Kernel Clustering (MKC) seeks to enhance clustering quality by optimally combining a set of given base kernels. Simple Multiple Kernel K-means (SMKKM) extends the supervised kernel alignment criterion to an unsupervised setting and has become a representative approach in MKC due to its strong clustering performance. Localized Simple Multiple Kernel K-means (L-SMKKM) further enhances clustering effectiveness by introducing a localized alignment criterion, addressing the limitation of SMKKM in handling intra-cluster variation of samples. However, the performance of L-SMKKM relies heavily on the neighborhood mask derived from the average kernel. Since neighborhoods based on the average kernel may not accurately reflect the intrinsic local structures of individual base kernels, applying a unified neighborhood mask across all kernels may cause mismatched local adaptations, leading to structural distortion and degraded clustering performance. To address this limitation, we propose a novel algorithm called Self-Neighborhood Kernel Simple Multiple Kernel K-means with Average Kernel Regularization (SNK-SMKKM-AKR). Our method constructs a self-neighborhood mask for each base kernel to preserve its local structure, enabling self-sampling to generate self-neighborhood kernels, which are then used in standard SMKKM. In addition, we incorporate average kernel alignment as a global regularization term to mitigate the effects of noise and local fluctuations. By integrating local adaptability with global consistency, our method improves clustering robustness and accuracy. Empirical evaluation on several benchmark datasets confirms the superior performance of our approach.</p>