A Lightweight Framework for Fast Image Retrieval on Large-Scale Image Datasets

Content-based image retrieval (CBIR) has attracted substantial attention over the past decade. Instead of taking textual keywords as input, CBIR techniques directly take a visual query and try to return its similar images from a given database. In this paper, we introduce a lightweight framework to fast locate the images on large-scale datasets. To achieve this, we first group the images with the high similarity into one cluster by using the deep neural network. Then, we extract the eigenvalues and eigenvectors from the images. Based on the eigenvalues and eigenvectors, we propose CB-tree (clustering binary tree) to fast locate the image clusters. Compared with the baseline schemes, the proposed framework can enhance the searching speed by up to 33%.