Dynamic Video Segmentation Method Based on Weighted Non-convex Regularization and Iteratively Reconstrained Low-Rank Representation

Abstract

A dynamic video segmentation method based on weighted non-convex regularization and iteratively re-constrained low-rank representation, including the following steps: (1) introduce a weighting factor W for the error matrix, and determine the constraint form of the weight matrix; (2) combine the W matrix to calculate the space Laplacian structure matrix L; (3) introduce weighted non-convex Rational function for the singular value of representation matrix Z; (4) optimize the existing GLRR framework through steps 1, 2 and 3, and propose the IRWNR model; (5) adopt the IRM framework (algorithm 1) Iteratively optimize the unknown variables W, L and Z in the target model; (6) Use the proximal gradient (EIPG) algorithm (Algorithm 2) to solve the subproblem of Z; (7) Use the block singular value threshold approximation method (Algorithm 3) Realize the SVT operation in Algorithm 2; (8) Iteratively optimize to obtain W, L and Z, which are used to realize dynamic video segmentation. It has the advantages of high operating efficiency, strong data adaptability, high accuracy and strong robustness.

Description

technical field [0001] The invention relates to a dynamic video segmentation based on weighted non-convex regularization and iterative heavy-constraint low-rank representation method. Background technique [0002] Among the methods for capturing subspace data and subspace data structure, low-rank representation is a promising method. Low-rank representations are widely used in the fields of signal processing and computer vision, such as scene classification, dynamic segmentation, face recognition and anomaly detection. The superiority of low-rank representation is mainly manifested in the following three aspects: first, the natural assumption of multiple low-order subspaces under the observation data; second, self-representation with specific anti-noise constraints; third, the use of nuclear norm low-rank Regularized convex approximation. However, these features are also where the limitations of low-rank representations lie, in that the structure of errors should be known ...

AI Technical Summary

Problems solved by technology

[0005] The invention solves the shortcomings of high-dimensional data calculation complexity, poor real-time performance, and poor anti-noise ability in the traditional low-rank representation method, and proposes a weighted non-convex regularized iterative reconstruction low-rank representation method for dynamic video segmentation

Images