基于自适应模糊C均值与后处理的图像分割算法

Image Segmentation Based on Adaptive Fuzzy C-Means and Post Processing Correction

由于图像噪声强度和边界的不确定性,图像分割算法的抗噪性和准确性是一项具有挑战性的任务,提出两种改进的模糊聚类算法用于图像分割。本文算法共分两步:第一步利用各像素邻域信息自适应地对中心像素进行噪声可能性检测,噪声与图像细节参数用以构建新的加权图像,结合新图像给出两种新颖的模糊聚类算法;第二步对分割结果中可能存在的错分点进行检测并对其进行后处理,从而提高分割准确度和视觉效果。在不同的噪声水平下,利用人工合成图像、Berkeley图像及其他图像对本文算法进行分割实验,结果表明,相比于其他模糊聚类算法,本文算法在分割准确率和ARI(Adjusted Rand Index)上具有优势,而且分割结果图像轮廓清晰,视觉效果更好。

Due to the image noise and boundary uncertainty, the noise resistance and accuracy of image segmentation algorithm is a challenging task. Two improvement fuzzy clustering algorithms for image segmentation are proposed. The proposed algorithms for image segmentation act as the following two steps. The first step is detecting the probability of every central pixel being a noise point adaptively based on the grey levels in its local information. The detecting results, playing the roles of denoising and detail information, are used to construct a new image, and then two novel segmentation algorithms based on fuzzy clustering are proposed. The second step is detecting the potentially misclassified pixels and refining the segmentation results by correcting the errors of clustering for improving the segmentation accuracy and visual effects. The obtained segmentation algorithms are carried out on synthetic image, Berkeley images and other real images in different noise levels. The results show that the proposed algorithm has advantages of segmentation accuracy and adjusted rand index compared with the others fuzzy clustering algorithms, and the segmentation results have clear contour and better visual effects.