Mobile augmented reality (MAR) systems utilize local features to represent structures such as corners or blobs in query videos and database images. Local features are compressed and sent to servers to perform query-database matching. The compression efficiency is important for the server-based MAR systems in terms of network latency. Conventional compression methods handle video frames independently. In recent years, an interframe compression technology, which exploits temporal correlation over

more »

... frames, has been proposed. It propagates local features temporally and realizes significant bitrate reduction without degradation in matching performance. However, this system requires a large amount of information when inserting independently compressed frames to update the status of local features. In this paper, we propose the selective update framework for local features and give simple algorithms to make use of the framework. Experimental results show that our proposal achieves gradual refreshes of local features, avoiding transmission of large amounts of data.