【摘要】 由于不同对象区域之间强度范围的重合,在存在强度不均匀的情况下很难分割图像。针对这一问题,提出一种新的水平集分割方法。在非均匀图像模型的基础上,推导出图像域的最优分割平面。在平面上,提出一种新的基于区域的压力函数,并在水平集公式中定义一个能量泛函。通过对能量泛函最小化,在对非均匀图像分割的同时,对偏置场进行估计。另外,为了准确估计偏置场,针对核函数设计一种新的自适应尺度参数。在真实图像和人工合成图像上的实验结果表明,该方法在精度、效率和鲁棒性方面有很好的优越性。更多还原

【Abstract】 Because of the overlap of intensity range between different object regions, it is difficult to segment the image when the intensity is not uniform. To solve this problem, a new level set segmentation method is proposed. Based on the non-uniform image model, the optimal segmentation plane of image domain was derived. On the plane, a new region-based pressure function was proposed, and an energy functional was defined in the level set formula. By minimizing the energy functional, the bias field was estimated while the non-uniform image was segmented. In addition, in order to estimate bias field accurately, a new adaptive scale paremeter was designed for kernel function. The experimental results on composite images and real images show the superiority of this method in accuracy, efficiency and robustness.更多还原