Performance Assessment of Determinant Kernels in Content-Based Video Retrieval: A Non-Square Determinant Approach

Abstract

Content-Based Video Retrieval (CBVR) relies on efficient similarity metrics to enhance the accuracy of the retrieval and reduce the computational overhead. This work explores the efficiency of determinant kernels by computing non-square determinants for similarity score calculation in high-dimensional feature spaces. Based on determinant-based computations, we propose a new method that increases the efficiency as well as the retrieval quality. The work focuses on the creation of a rigorous mathematical framework that integrates determinant kernels in the similarity measurement process, which forms a crucial building block in CBVR systems. Based on rigorous experiments on standard video datasets, we analyze the degree to which non-square determinant computations capture variations in the video content that other approaches might not capture. Our analysis demonstrates that the new method significantly increases the processing efficiency, leading to more accurate retrieval results. Additionally, the method proves to be computationally efficient, making it suitable for real-time applications as well as for large-scale video databases. The findings of this work make substantial contributions to the area of CBVR techniques and pave the way for future work in efficient video content analysis.