Tensor structural missing filling method based on joint low rank and sparse representation

Abstract

The invention relates to the field of computer vision. In order to propose a tensor structural missing filling method and realize the accurate filling of the structural missing tensor, the tensor structural missing filling method based on the combination of low rank and sparse representation and the TT low rank tensor filling theory are introduced into the TT low rank prior to constrain the potential tensor. At the same time, considering that the fiber signal along each dimension of the tensor can be sparsely represented by a dictionary, and the missing fiber of the previous dimension can be recovered by sparse constraint on the fiber signal in the next dimension, sparse constraint is introduced for each dimension of the fiber signal; based on the sparse apriori of the combined TT low rankand each dimension, the tensor filling problem with structural defects is formulated as a constrained optimization problem, so that the tensor filling problem with structural defects can be realized.The invention is mainly applied to video image inpainting, recommendation system, data mining and multi-classification learning occasions.

Description

technical field [0001] The invention belongs to the field of computer vision, in particular to a tensor structural missing filling method based on low-rank tensor filling and sparse representation theory. Background technique [0002] Tensor padding is to restore unknown missing elements based on some known elements of tensors. This problem has attracted extensive research and attention in recent years. This kind of data filling problem is often encountered due to the incompleteness of data in many fields of computer vision and machine learning, such as video image restoration, recommendation system, data mining and multi-classification learning. [0003] In recent years, there have been a lot of research results on methods to solve the tensor filling problem. Due to the ill-conditioned nature of the tensor filling problem, the current tensor filling method generally assumes that the tensor to be restored is low-rank or approximately low-rank, and then obtains the missing e...

AI Technical Summary

Problems solved by technology

[0004] At present, in view of the above-mentioned structural missing situation, the academic community only considers this structural missing in two-dimensional matrices. In high-dimensional tensor data, only random missing is still generally considered, and most of them only use tensors. Thaker's low-rank feature is used for constraint solving, which cannot effectively recover tensor data with structural missing

Images