A Hashing Image Retrieval Method Based on Deep Learning and Local Feature Fusion

The multimedia information such as images and videos has been growing rapidly, how to efficiently retrieve large-scale image dataset to meet user needs is an urgent problem. The traditional method has the problem of slow retrieval and low accuracy on large-scale datasets, we propose an effective deep learning framework to generate binary hash codes for fast image retrieval, our idea is to fuse local features maps of different layers in convolutional neural networks (CNNs), and the binary hash codes can be learned by employing a hidden layer. Additionally, we train the network by combining cross entropy loss function with the triplet loss function to get better features. The approximate nearest neighbor search strategy is used to improve the quality and speed of retrieval. Experimental results show that our method outperforms several state-of-the-art hashing image retrieval algorithms on the MNIST and CIFAR-10 datasets. At last, we further demonstrate its scalability and efficacy on the CUB200-2011 and Stanford Dogs fine-grained classification datasets.