Multi-Query Image Retrieval using CNN and SIFT Features

Due to the rapid growth of image number, the content-based image retrieval becomes an indispensable tool for huge database. In this study, our focus is retrieving a specific building at different viewing angles stored in a database. In addition, if the user can provide additional images as the second and/or the third queries, how do we combine the information provided by these multiple queries? Thus, we develop a multi-query fusion method to achieve a higher accuracy. Although Deep Neural Net (DNN) can provide an End-to-End image retrieval system, we like to see if the traditional image feature can offer additional performance improvement. That is, we test two different types of features designed for image retrieval purpose. We adopt the Scale-Invariant Feature Trans form (SIFT) features as the low-level feature and the Convolutional Neural Network (CNN) features as the high-level feature in the retrieval process. The AlexNet is used as our CNN model and also, its extension to the Siamese-Triplet Network is in use to match the image retrieval purpose. Several data fusion structures haw been prop:wed Our best system exceeds most of the state-of-the-art retrieval methods for a single query. The multi-query retrieval can further increase the retrieval accuracy, which is rarely studied by the other researchers.