自适应灰度加权的鲁棒模糊C均值图像分割

针对传统模糊C均值(fuzzy C-means，FCM)算法以及结合空间信息的相关改进算法分割精度较低、对噪声敏感的问题，提出一种自适应灰度加权的鲁棒模糊C均值图像分割算法。首先，通过定义像素间的局部灰度相似性测度来反映各像素对局部邻域的影响程度，并根据邻域窗口中各像素的灰度差异，利用指数函数进一步控制邻域像素的影响权重，实现像素灰度的自适应加权，从而提高像素灰度计算的准确性。其次，构造出一种改进的距离测度代替传统的欧氏距离，用于计算各像素与聚类中心之间的相似距离，增强算法对噪声和异常值的鲁棒性。最后，将提出的自适应灰度加权方法与改进的距离测度应用到FCM算法中，实现图像分割。实验结果表明，该算法需根据图像噪声的强度适当地选取邻域窗口大小，在此条件下算法能够取得较优的分割效果和运行效率，且对噪声具有较强的鲁棒性。

Adaptive gray-weighted robust fuzzy C-means algorithm for image segmentation

The traditional fuzzy C-means (FCM) algorithm and its corresponding improved algorithm that is combined with spatial information have low segmentation accuracy and poor robustness to noise. To address these defects, we propose a robust FCM image segmentation algorithm based on adaptive gray-weighting. First, we define a local grayscale similarity measure for pixels to reflect the influence of all pixels on the local neighborhood. Regarding the grayscale difference between pixels in a neighborhood window, we utilize an exponential function to further control the influence weight of a neighborhood pixel and realize adaptive weighting of the pixel grayscale to improve its calculation accuracy Next, to strengthen the robustness of the algorithm to noise and outliers, we use an improved distance measure to replace the traditional Euclidean distance and use it to calculate the similarity distance between the pixels and the clustering center. Finally, we apply this new method based on adaptive gray weight and enhanced distance measurement to an FCM algorithm for image segmentation. Our experimental results show that, for the algorithm, the size of the neighborhood window must be properly selected on basis of the noise intensity of an image. Under this condition, an excellent segmentation effect and operational efficiency can be achieved, in addition to excellent robustness to noise.