The Density Peak Clustering (DPC) is a widely recognized algorithm known for its intuitive principles and ability to detect non-spherical clusters. However, it suffers from significant drawbacks, including high sensitivity to parameters and difficulty in identifying cluster centers within low-density regions. To overcome these limitations, this paper introduces a novel variant named Density Peaks Clustering based on Mutual Nearest Neighbor Density Allocation (DPC-MNND). The proposed algorithm incorporates a kernel-based density estimation method using Reverse K-Nearest Neighbors (RNN), which eliminates reliance on the cut-off distance and enhances adaptability to multi-density datasets. Furthermore, the RNN mechanism is integrated into the cluster center selection process, allowing more accurate identification of potential centers across density variations. A two-phase label propagation strategy is also designed to improve robustness and reduce error propagation during cluster assignment. Comprehensive experiments were conducted on a diverse set of synthetic and real-world benchmarks. The results demonstrate that DPC-MNND consistently outperforms or matches state-of-the-art clustering methods, including DPC, DPC-DBFN, HDBSCAN, and Spectral Clustering, across multiple evaluation metrics such as ACC, ARI, AMI, NMI, and FMI. These findings confirm that DPC-MNND offers superior effectiveness and robustness in handling complex data structures with varying densities.