Abstract
For large-scale image retrieval, high-dimensional image representations derived from pre-trained Convolutional Neural Networks (CNNs) make the retrieval system inefficiency. In this paper, we propose to combine nonlinear dimension reduction and hashing method for efficient image retrieval. We firstly extract 4096-dimension features by a pre-trained CNNs model. Secondly, we use t-Distributed Stochastic NeighborEmbedding (t-SNE) to reduce the deep features to 1024-dimension. Finally, we use Sparse Projection (SP) to build 256 bits binary encoding features for image retrieval. We assess the performance of our proposed method with the Oxford5k, Paris6k and Holidaysdatasets. The experimental results show that the proposed method can achieve the state-of-the-art accuracy on common benchmarks.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call