Multi-view subspace clustering network with block diagonal and diverse representation.

•Multiple deep autoencoders are established to learn the more discriminative feature space for the multi-view data.•A-block diagonal regularizer is imposed on the weight of the self-expression module, seeking the block diagonal representation matrix directly.•A diverse representation is introduced into this subspace clustering network to reduce the redundancy, obtaining more diverse representation matrices.•Experimental results obtained on the UCI Digit, Caltech101-20, COIL100 and Caltech101-7 datasets have demonstrated the superior performance.