[This article belongs to Volume - 53, Issue - 05]
Gongcheng Kexue Yu Jishu/Advanced Engineering Science
Journal ID : AES-14-11-2021-79

Title : Improved Truncated Nuclear Norm and Its Application in Video Foreground-background Separation
YANG Yongpeng, YANG Zhenzhen, LI Jianlin, FAN Lu,

Abstract :

The main purpose of video background separation is to extract objects of interest from the video, but it is still one of the most challenging tasks in computer vision and other fields due to the influence of noise and lighting changes. The truncated nuclear norm (TNN) algorithm is a classic robust principal component analysis (RPCA) algorithm, which is widely used to separate the background and the front of the video. However, the truncated kernel norm in this algorithm does not have a high degree of approximation to the rank function in the traditional robust principal component analysis, resulting in poor stability and low accuracy in separating the front and background of the video in some complex scenes. To solve this problem, this paper proposes an improved truncated nuclear norm (improved truncated nuclear norm, ITNN) algorithm. The algorithm first replaces the kernel norm in the TNN model with a non-convex γ norm, and analyzes that the non-convex γ norm has a higher degree of approximation to the rank function than the kernel norm . Corresponding model; secondly, in order to solve the proposed model, this paper introduces the generalized alternating direction method of multipliers (GADMM) to solve the model; finally, the proposed ITNN algorithm is applied to multiple public videos In the previous background separation experiment, and by showing the foreground effects of different videos, the effectiveness of the ITNN algorithm was verified from a visual point of view. At the same time, the F-measure value of the video foreground extracted by the proposed algorithm and the comparison algorithm is calculated, which further verifies the effectiveness of the ITNN algorithm from the perspective of quantification. In addition, the experiment also recorded the running time of the video front and background separation of each algorithm, which verified the efficiency of the ITNN algorithm. In a word, this paper verifies the effectiveness and superiority of the proposed ITNN algorithm in the separation of video front and background through experiments.