A Local Taylor Fitting Level Set Image Segmentation Model With Global Region-based Information

Abstract

The traditional CV model based on global information has strong robustness to the initial contour and noise, while the Taylor expansion fitting model (LATE) based on local information can segment the image with intensity inhomogeneity, but is sensitive to the initial contour and noise. To solve the above problems, an adaptive level set image segmentation combining global (CV) and local information (LATE) was proposed. Firstly, the global and local energy terms were constructed respectively by using the global and local information of the image, and a weight coefficient was constructed between the global and local energy terms for adaptive adjustment. Secondly, the energy penalty term and length term are added to avoid the reinitialization of the level set function and smooth the curve evolution, thus improving the stability of numerical calculation. In the end, some comparative experiments show that the proposed model is robust to initial contour and noise, and has higher segmentation accuracy and efficiency for non-uniform gray images and natural images.