Tree Histogram Coding for Mobile Image Matching

摘要: For mobile image matching applications, a device captures query image, extracts descriptive features, and transmits these features wirelessly to server. The server recognizes the by comparing extracted its database returns information associated with recognition result. slow links, feature compression is crucial for low-latency retrieval. Previous retrieval systems transmit compressed descriptors, which well suited pairwise matching. fast from large databases, however, scalable vocabulary trees are commonly employed. In this paper, we propose rate-efficient codec designed tree-based By encoding tree histogram, our can achieve more than 5x rate reduction compared sending descriptors. discarding order amongst list of histogram coding requires 1.5x lower node index every feature. A statistical analysis performed study how entropy encoded symbols varies depth number features.