Active Contour Model Based on Fuzzy C-Means and Local Pre-fitting Energy for Image Segmentation

Abstract Active contour models (ACM) have achieved remarkable results in image segmentation. However,existing ACMs have some shortcomings such as over-dependence on the initial contour, complexparameter adjustments, and difficulty in balancing segmentation accuracy and speed. To furtherimprove the performance of ACM, this paper proposes an active contour model based on fuzzy c-means and local pre-fitting (FLPF) function. Firstly, a linear weighted image is defined as a samplefor the fuzzy c-means (FCM) clustering algorithm to pre-fit the image intensity. A pre-processingoperation is proposed to improve the FCM clustering algorithm and increase the computation speed.Then, second-order differential data-driven terms are designed to guide the curve evolution rapidlyand adaptively toward the target boundary. In addition, adaptive regularization functions are constructed to optimize and normalize the data-driven terms and level set functions, which improves therobustness of the proposed model. Finally, an improved parameter tuning framework based on thedeep learning algorithm YOLOv5 is proposed for the FLPF model to achieve automated parameteradjustments. Compared to other 6 models, our model has advantages in segmentation speed and accuracy, decreasing the average segmentation time by 77.5% and improving the average segmentationaccuracy by more than 7.6% of 10 images.