Kernel Subspace Clustering with Block Diagonal Prior

We present in this paper a kernel subspace clustering with block diagonal prior (KSCBD) method for nonlinear data analysis. Unlike most existing block diagonal constrained subspace clustering methods can only be performed explicitly on the original data space, our KSCBD embed the block diagonal prior into the kernel Hilbert space. Specifically, our KSCBD first projects the data to a high-dimension linear space by kernel method. Then, the block diagonal prior is embedded to the kernel Hilbert space for keeping the block diagonal structure. Experimental results on both synthetic and real-world data sets demonstrate the effectiveness of the proposed algorithm.