Research on Feature Dimensionality Reduction in Content Based Public Cultural Video Retrieval

Content features and feature dimensionality reduction play an important role in content-based public cultural video retrieval. The high dimensionality of content features will affect the efficiency of video retrieval. Feature dimensionality reduction is to reduce the dimension of the feature vector as much as possible, that is, to remove the redundancy between features, and to express most of the information with a small amount of data. In this paper, combining the characteristics of public cultural videos, we use the deep learning framework Caffe combined with AlexNet network model to extract the content features of public cultural videos and principle component analysis method to reduce the dimensionality of public cultural video content features. We use these methods to perform retrieval experiments in the public cultural dataset and analyze the results. Experiments show that this dimension reduction method has high retrieval efficiency for public cultural videos.