Quality-aware Mobile Visual Search

With the increasing popularity of mobile phones and tablets, mobile visual search has attracted growing interest in the field of content-based image retrieval (CBIR). In this paper, we present a novel framework for quality-aware CBIR. On the mobile- client side, a query image is compressed to a certain quality level to accommodate the network condition and then uploaded onto a server with its quality level transferred as side information. On the sever side, a set of features are extracted from the query image and then compared against the features of the images in the database. As the efficacy of different features changes over query quality, we leverage the side information about the query quality to select a quality-specific similarity function that is learnt offline using a Support Vector Machine (SVM) method. The experimental results have demonstrated the potential of our framework.