A Self-Developed GVFsnake Image Segmentation Method Based on Local Contrast Enhancement

In the traditional GVFsnake (Gradient Vector Flow active contour model) method, while the algorithm proposed by Xu et al. effectively handles concave boundary issues, its segmentation performance remains limited for images with low contrast or severe noise interference. Addressing this critical challenge, this paper presents an improved GVFsnake method combined with local contrast enhancement.

The core innovation of this method lies in the preprocessing stage: first performing adaptive local contrast enhancement on the input image. This is implemented by analyzing pixel neighborhood characteristics to dynamically adjust enhancement intensity, effectively highlighting gradient information in weak edge regions. This preprocessing enables subsequent GVF energy field calculations to capture more genuine edge features, particularly suitable for scenarios with blurred tissue boundaries in medical imaging. The implementation typically involves calculating local statistics (mean and standard deviation) within sliding windows and applying contrast-limited adaptive histogram equalization (CLAHE) with region-specific parameters.

In terms of energy function design, the improved method preserves the advantages of the original GVFsnake's external energy term while optimizing the weight distribution strategy for internal energy terms. By introducing an edge saliency detection mechanism, the algorithm automatically adjusts the evolution speed of contour points in different regions—accelerating convergence in strong edge areas while enhancing exploration capability in weak edge regions. This is achieved through a weighting function based on gradient magnitude thresholds and regional homogeneity analysis, implemented via conditional statements in the contour evolution loop.

Experimental results demonstrate that while maintaining the topological adaptability of traditional GVFsnake, this method significantly improves segmentation accuracy for intensity inhomogeneity regions in CT/MRI images. It exhibits particularly strong robustness in tumor boundary localization and vascular tree structure extraction tasks. This enhancement allows the algorithm to better meet the precision requirements of modern medical image analysis while maintaining computational efficiency, with optimized matrix operations for GVF field computation and parallel processing for contour point updates.