<p>This paper proposes an unsupervised intrinsic landmark detection method that enhances non-rigid dense shape correspondence. Our approach identifies intrinsic landmarks on near-isometric surfaces, enabling efficient and robust point-wise correspondence. A key observation is that a small set of intrinsic landmarks is sufficient to reconstruct an isometric map, allowing for effective shape matching. Our method eliminates the need for manually defined landmarks by leveraging intrinsic geometric properties. Specifically, we compute the Karcher means of the surface and sequentially select the farthest points from these medians. This provides a stable and repeatable landmark set, which can be used in various shape correspondence frameworks. We evaluate our approach on multiple shape matching benchmarks and show that it consistently improves the accuracy of existing correspondence methods without requiring modifications. The results demonstrate that our method is computationally efficient and can serve as a plug-and-play component in shape matching pipelines. Code is available at <a href="https://github.com/ZHLRJ/Non-Rigid-Shape-Correspondence">https://github.com/ZHLRJ/Non-Rigid-Shape-Correspondence</a>.</p>