Image retrieval based on convolutional neural network and kernel-based supervised hashing

With the increasing amount of image data, the present image retrieval methods have several problems, such as the steps of the visual features coding are fixed, lack of learning ability, low expression ability of the features, high dimension of the features, which restrict the retrieval performance severely. Aiming at these problems, an image retrieval method based on convolutional neural network and kernel-based supervised hashing is proposed. Firstly, we use the learning ability of convolutional neural network to mine the internal implication relation of the images and extract the deep features. Then, introduce the kernel-based supervised hashing and train the high-dimension deep features with the supervised information, map the high-dimensional features to the low-dimensional compact binary codes. Finally, image retrieval on the mass image datasets is accomplished effectively in low-dimensional hamming space. The experimental results on ImageNet-1000 and Caltech-256 demonstrate that our method can enhance the expression ability of the image features effectively, and reduce the dimensionality of the high-dimension image features, the image retrieval performance is superior to the state-of-the-art methods.