Non-linear Statistical Filter Fusions for Optimal Intensity Fitting in Active Contour-Based Image Segmentation

Active Contour Models (ACMs) are renowned for their parameter diversity and tunability, with each parameter carrying significant meaning that greatly influences image segmentation. The fitting parameters, crucial for the segmentation process, are traditionally iteratively calculated, incurring substantial computational costs. To address this challenge, we introduce an approach in this study, utilizing prefitting terms. These special fitting parameters are determined before the level set iteration, aiming to reduce the computational burden associated with fitting parameter calculations. After calculating the prefitting terms, these are incorporated in a quadratic loss function. Our proposed method involves the computation of pre-fitting terms through the average of median and either maximum or minimum(order statistic terms) values in a local window. Directly using min-max filters may impact salt and pepper noise. By avoiding direct utilization of min-max filters, our approach mitigates the problem of salt and pepper noise. Additionally, incorporating the median aids in disregarding outlier noise within the model. Finally our research findings indicate that the model demonstrates strong initial robustness, along with enhanced speed and reduced iteration requirements. Moreover, it yields superior Dice scores and Jaccard indices for segmenting natural images.