A fast level set segmentation algorithm for sonar images driven by Gaussian Mixture Model probability

Sonar images are frequently degraded by speckle noise due to the inherent scattering properties of acoustic signals, leading to blurred edges, low contrast, and uneven grayscale distributions. These characteristics substantially complicate sonar image segmentation. To mitigate these challenges, this paper proposes a fast level set segmentation algorithm driven by the probability of the Gaussian Mixture Model. The Gaussian Mixture Model extracts grayscale statistical features from distinct regions to formulate the fidelity term in the Chan–Vese model. A novel posterior probability-based level set function is introduced to establish bidirectional coupling between the Gaussian Mixture Model and the Chan–Vese model. In this framework, the Gaussian Mixture Model first guides rapid level set evolution by propagating its state to the Chan–Vese model. Subsequently, during energy minimization, the Gaussian Mixture Model receives feedback information for the dynamic estimation of the geometric-prior distribution, which includes spatial and curvature information, thereby enhancing edge detection accuracy. Additionally, morphological operations are integrated into the iterative process to approximate posterior probabilities, effectively suppressing isolated noise and repairing discontinuities. The synergistic interaction between the two models enables efficient sonar image segmentation. Experimental results demonstrate that the proposed algorithm outperforms mainstream methods, achieving over a fivefold improvement in segmentation efficiency.