High-performance image retrieval based on bitrate allocation

Image retrieval plays a key role in finding the required online multimedia learning materials quickly. In this paper, we realize a high-performance image retrieval by using bitrate allocation. First, the global descriptor is extracted to represent each image of the database. Second, the intra-group distance and the inter-group distance for each image are computed based on the global descriptor. Then, the retrieval difficulty degree (RDD) is evaluated based on the two types of distances. After that, an accurate bitrate allocation scheme is proposed when building the feature database according to RDD: using a higher bitrate for images that are hard to retrieve and lower bitrate for simpler ones. Finally, the higher retrieval performance is achieved based on the constructed highly efficient database. Experiments demonstrate that our method far outperforms the state-of-the-art on image retrieval benchmarks.