Color Image Retrieval Based on Quaternion and Deep Features

In this paper, aiming at the inefficient question of feature extraction and multi-feature fusion in existing algorithms, we propose a color image retrieval algorithm based on quaternion and deep features. Firstly, we replace the real part of the quaternion with the image information entropy to form a new image matrix, and extracting its moment invariant as the traditional features. Meanwhile, the deep features are extracted by the improved convolutional neural network. Secondly, we develop an effective features fusion procedure. Finally, retrieval results are obtained by Euclidean distance between image's fusion features. In order to demonstrate the effectiveness of our method, we have conducted extensive experiments on three well known datasets (INRIA Holidays dataset, University of Kentucky Benchmark, Oxford 5k).