Compressing Visual Descriptors of Image Sequences

In recent years, there has been significant progress in developing more compact visual descriptors, typically by aggregating local descriptors. However, all these methods are descriptors for still images, and are typically applied independently to (key) frames when used in tasks such as instance search in video. Thus, they do not make use of the temporal redundancy of the video, which has negative impacts on the descriptor size and the matching complexity. We propose a compressed descriptor for image sequences, which encodes a segment of video using a single descriptor. The proposed approach is a framework that can be used with different local descriptors, including compact descriptors. We describe the extraction and matching process for the descriptor and provide evaluation results on a large video data set.