Multiple feature similarity based for image retrieval

In Bag-of-Words-based image retrieval, the local feature could not describe the global information of an image. It produces many false matches and reduces the retrieval precision. To address this problem, this paper proposes a new method which is based on the global and local feature similarity. The global feature extracted by convolutional neural network is added to the local keypoints extracted in a given image. The local and global features are used together to improve the accuracy and a 2-D inverted index is built to accelerate the speed of retrieval. Experimental results demonstrate that the method proposed in this paper can improve the accuracy significantly. It outperforms the state-of-the-arts on the Ukbench and Holidays datasets.