The Wind Power Curve (WPC) reflects the operating condition of its components to a certain extent, which is of great significance in guiding the operation and maintenance of wind farms. Previous curve modeling studies have focused on modeling based on wind turbine operational data and have resulted in the development of two main modeling approaches, parametric and nonparametric. In this paper, a new image-based power curve modeling algorithm is introduced to convert the modeling problem into the problem of extracting the skeleton of WPC image. Specifically, the raw data is cleaned using image processing methods, clustering is performed based on a Gaussian Mixture Model (GMM) to generate density maps. Subsequently, the Fuzzy Distance Transform (FDT) is applied to extract the skeleton of the wind speed-power images as the initial power curve, which is smoothed using a fitting function. Experiments on data from 17 wind turbines in actual operation demonstrate the effectiveness of the proposed method in the power curve modeling task.