Post Processing Algorithm for Background Subtraction Model based on Entropy Approximation and Style Transfer Neural Network
Tin-Thai Trung, Synh Viet‐Uyen Ha · 2022 RIVF International Conference on Computing and Communication Technologies (RIVF) · 2022
Background subtraction which aims to detect foreground masks is a fundamental task in surveillance systems. On-going research in background subtraction models has shown that background modelings are a highly robust technique to extract regions of interest, known as foregrounds. However, these background modeling techniques still failed to reproduce backgrounds correctly. Therefore, our research focuses on background correction by filling the missing information in the reproduced background image. In this paper, we proposed a post-processing method to correct a constructed background after a novel framework of a tensor-driven GMM for background modeling. Our proposed method which is based on the entropy approximation not only improves the F-score evaluation of the TensorMOG modeling but also can be competitive with state-of-the-art models in terms of processing time.