为解决由于自然纹理的干扰而导致的分割图像边缘模糊问题,对模糊C均值聚类算法进行改进并应用于交互式图像分割中。用户通过输入种子点来获得目标和背景的主要特征,并将输入的种子点作为聚类中心点;提出全局空间相似性度量标准并引入Gabor能量滤波器来计算图像中各点到聚类中心的距离;算法首次引入边缘密度概念定义权重因子,根据图像特点,自适应地计算图像中任意一点的纹理特征和颜色特征在特征空间中所占比例,使得到的特征更加准确地描述图像的本质属性。对具有自然纹理背景的图像进行仿真实验,应用两种性能指标来比较本文所提算法与随机游走算法的分割精度。实验结果表明,本文算法分割精度高于模糊聚类和随机游走算法。

We propose an improved(FCM)approach for interactive image segmentation to solve the problem that the objective contour is easily influenced by natural texture.Users interactively input seed points as clusters centers to get the main feature differences between foreground and background;a global spatial similarity measure model and Gabor energy filters are introduced into the distance metric to measure the similarity between all the pixels and the cluster centers.Edge density is introduced into the definition of the weight factor,according to image features,the ratio of the texture feature and the color feature in the feature space for pixels which can be adaptively calculated,making the feature we calculated more accurately reflect the essential attributes of the image.In this paper,experiments have been carried out on images with natural texture background,and two performance measures were used to compare the proposed algorithm with the random walker(RW)algorithm.It is shown that the accuracy of the proposed method surpasses FCM and RW methods.