基于KL变换的模糊C-均值聚类彩色图像分割

根据图像色彩特征空间的正交特性，以及构成特征空间的特征向量和特征值之间的统计特性，提出了一种新的彩色图像指定区域分割算法。首先在指定区域选取采样像素，通过KL变换计算采样像素的协方差矩阵、特征值、特征向量；由特征向量构成指定区域的色彩特征空间，然后对原色彩空间中的向量进行空间变换和权重变换；最后用模糊C-均值聚类方法聚类变换后的向量，得到分割结果。文中给出了静物图像的聚类分割结果，体现了算法对于指定区域细节分割的准确性。

Fuzzy C-Clustering Based on KL Transform for Color Image Segmentation

On the basis of the orthogonality and the statistical prosperities between eigenvectors and eigenvalues of the color eigenspace, a new algorithm to accurately segment the designed images that have the same color as the pre-selected color pixels is proposed. At first, correlation matrix, eigenvalues and eigenvectors are computed through KL transforming pre-selected sample pixels. Hence, eigenspace that takes eigenvectors as the coordinates is formed. Then apply the eigenspace-transform and weight-transform to the color pixel vectors belong to the normal color space. At last, the transformed pixel vectors are clustered by FCM and the segmented result is obtained. Experiments show the method can effectively achieve the desired color image segmentation.