Particle swarm optimized deep spatio-temporal features for efficient video retrieval

In content-based video retrieval, the phases of video frame selection and 3-dimensional feature extraction are especially crucial. These stages should be optimized based on temporal complexity because increasing the execution time would also increase the retrieval time. Furthermore, information optimization is not taken into consideration by fixed keyframe sampling methods such as clustering, uniform sampling, and shot boundary-based, which might result in information redundancy in the selected keyframes in the video data. To address these problems, a deep spatiotemporal feature extraction method based on particle swarm optimization has been proposed for efficient video retrieval. The results demonstrate that the approach outperforms several benchmark sampling methodologies. The optimization is based on the grayscale histogram. Deep features may potentially serve as the foundation for the next iterations of this.