An active contour model based on Kullback–Leibler divergence and morphology for image segmentation with edge leakage

Intensity inhomogeneity and edge leakage tend to make traditional image segmentation methods unavailable due to uneven lighting and severe noise, etc. Most active contour models (ACMs) perform ineffectively when utilized for these images with noise and intensity inhomogeneity. To relieve these issues, the robust ACM based on Kullback–Leibler divergence and mathematical morphology (KLMM) is proposed to segment images with intensity inhomogeneity. This Kullback–Leibler divergence is applied for the local energy term to differentiate intensity variation between the true distribution and fitting distribution in a local region, which integrates Retinex modeling. The fitting bias formulation is computed with morphological operators to correct intensity fluctuation variance and mitigate edge leakage caused by noise and uneven lighting. The data optimization function and the average filtering diminish accumulation errors in iteration and assure correct evolution. The experimental results on various real and synthetic images manifest the proposed model with its mIOU converging over 0.9 outperforms comprehensively several existing state-of-art models in accuracy and robustness.