TRÍCH CHỌN ĐẶC TRƯNG TỪ CÁC MẠNG NƠ-RON TÍCH CHẬP CHO BÀI TOÁN PHÁT HIỆN SAO CHÉP MỘT PHẦN VIDEO

Abstract

2D convolutional neural networks are the key component in partial video copy detection systems. They play a crucial role in video retrieval and matching tasks within a large database. However, the performance characteristics of these feature extraction methods have been little discussed in the literature. This paper presents two key contributions. First, we conduct the experiments on a large-scale dataset to demonstrate the generalization capability and clarify the performance characteristics of popular neural networks. Next, we propose a time-series model approach to highlight the advantages and limitations of image features extracted from neural networks in the partial video copy detection problem.