Finding relevant instances in databases has always been a challenging task. Recently a new method, called
Sparse Modeling Representative Selection
(SMRS) has been proposed in this area and is based on data self-representation. SMRS estimates a matrix of coefficients by minimizing a reconstruction error and a regularization term on these coefficients using the
matrix norm. In this paper, we propose another alternative of coding based on a two stage Collaborative Neighbor Representation in which a non-dense matrix of coefficients is estimated without invoking any explicit sparse coding. Experiments are conducted on summarizing a video movie and on summarizing training face datasets used for face recognition. These experiments showed that the proposed method can outperform the state-of-the art methods.