Now lots of researches on fuzzy kernel clustering for segmentation have been done. But the selection of the kernel parameter is not solved effectively. The result of segmentation is influenced by the kernel parameter

so that the segmentation is not adaptive and well. The use of hybrid kernels with pairwise constraints is proposed to the above problems. A composite gaussian kernel

which is composited by many gaussian kernels with different kernel parameter

is used to replace the traditional gaussian kernel. The information about classifications with pairwise constraints is to optimize the parameters of such a composite gaussian kernel. Then the composite gaussian kernel is used to cluster for segmentation. The experimental results demonstrate that this technique can solve the problem of selection of kernel parameter successfully and make the segmentation adaptively. The results of segmentation are robust and accurate using the hybrid kernels.