UNCORRELATED DISCRIMINATIVE LOWRANK PRESERVING PROJECTION FOR DIMENSIONALITY REDUCTION

Low-rank representation (LRR) is a tool which is used to ensnare the inherent representation of the noticed samples. Even LRR make an effective representation, but which cannot make better classification and which is weak while handling new samples and which cannot acquire projection matrix in the training period. A novel uncorrelated discriminative low rank preserving projection (UDLRPP) is introduced as dimension reduction algorithm by assimilating the un-correlation constraint and the internal region relationship of the original samples into the low rank representation (LRR). The role of UDLRPP is, the LRR can ensnare the overall structure data and the internal geometrical structure data is parallelly preserved by manifold regularization term. The low-rank representation coefficients are used to acquire the constrained term is instigated by the adaptive graph. And by adding the un-correlation analysis constraint expression, UDLRPP can attain the foremost projection matrix to enhance the classification accuracy. Several investigations completed on different public image datasets and declared that the proposed UDLRPP can acquire improved recognition rate compared with the state-of-the-art feature extraction methods.