<p>Considering the shortcomings of possibilistic C-means clustering in terms of anti-noise robustness and clustering consistency, this paper proposes a new robust kernelized spatial possibilistic log-local information C-means clustering with dual weighting exponents for image segmentation. First, this paper constructs a new log-local information factor and incorporates it into the possibilistic C-means clustering to improve anti-noise robustness; Secondly, the local spatial information of the current clustered pixel is used to construct a weighting factor, and the local typicality values of the current clustered pixel are weighted and combined to obtain a prior probability that achieves the distance of the current pixel from the cluster centre in the possibilistic C-means clustering constraints to further improve the noise robustness of the algorithm; Again, the kernel metric and double-weighting exponents are introduced into the spatial possibilistic C-means clustering with local information to form the proposed algorithm. Finally, the convergence of the proposed algorithm is strictly analyzed using Zangwill’s theorem. The experimental results show that the proposed algorithm outperforms existing possibilistic clustering algorithms in highly noisy images, with better segmentation performance and anti-noise robustness. Specifically, its PSNR and ACC values improved by approximately 0.01–6.74 and 0.01–0.05, respectively.</p>