【摘要】 为了改善传统基于聚类的图像分割算法对噪声敏感以及仅使用单一特征无法精确描述目标特性等问题,提出了一种基于区域的多特征图像分割算法。首先,使用Meanshift算法对原图像进行预分割,获得一组区域块;其次,提取每个区域块的颜色特征和纹理特征,使用FCM算法分别对每个特征进行聚类,针对每个特征获得一个类标签邻接矩阵;再次,将多个邻接矩阵叠加,形成多特征邻接矩阵;最后,使用NCUT算法对叠加邻接矩阵进行聚类,获得最终分割图像。实验结果表明,基于区域多特征的分割算法优于对比算法,融合多特征对图像分割可以更准确地识别不同的目标结构,具有更好的分割效果。更多还原

【Abstract】 In order to improve the sensitivity of traditional clustering-based image segmentation algorithm to noise and the inability to accurately describe the target characteristics by using only a single feature,a region-based multi-feature image segmentation algorithm is proposed. Firstly,the original image is pre-segmented by Meanshift algorithm,and a group of regions is obtained quickly. Secondly,the color and texture features of each block are extracted. FCM algorithm is used to split each feature separately,and a class label adjacency matrix is obtained for each feature. Thirdly,multiple adjacency matrices are superposed to form multi-feature adjacency matrices. Finally,the NCUT algorithm is used to split the superimposed adjacency matrix to obtain the final segmentation image. The experimental results show that the segmentation algorithm based on region multi-feature is better than the contrast algorithm. Fusion of multi-features for image segmentation can identify different target structures more accurately,and has better segmentation effect.更多还原