Latent Smooth Representation Segmentation with Structural Constraints

Abstract:
The latent subspace structure of high-dimensional data can be obtained through subspace clustering, and the block diagonal representation (BDR) model can effectively cluster data by using linear representation. Compared with other subspace clustering methods, the algorithm based on spectral clustering has received extensive attention because it can show better performance in many practical applications. The algorithm proposed in this paper aims at the reconstructed coefficient vector obtained by the sparse representation algorithm based on block diagonal structure (BDSR), and then uses the smooth representation algorithm (SMR) to process the reconstructed sparse vector, and a novel algorithm called latent smooth representation clustering with structural constraints (LSMRS) is proposed. According to the original data set, it can be well represented by the reconstructed coefficient vector, this algorithm can be regarded as executing SMR in the latent subspace found by BDSR, and hopes to obtain better performance. Finally, experiments on a famous database show that the algorithm is better than the related algorithms.