Image Hashing by Pre-Trained Deep Neural Network
Pingyuan Li, Dan Zhang, Yuan Xiaoguang, Jiang Suiping · 2022 Asia Conference on Algorithms, Computing and Machine Learning (CACML) · 2022
Image hashing can map a high-dimensional image to a low-dimensional binary code, so that it is widely used in image retrieval in big data. An image hashing method based on deep neural network is presented in this paper by expanding the usage of deep neural network. Firstly, a pre-trained deep neural network for image classification is used to extract the feature vector of the image. Secondly, the feature vector is transformed by discrete cosine transform to get low-frequency components. Finally, the low-frequency components are binarized to generate hash code of the image. Compared with commonly used perceptual hash, the proposed method keeps the generality of perceptual hash and is robust to image flipping and rotation. The experiments conducted on caltech-256 data set showed that the proposed method improved the accuracy of image retrieval and the quality of hash code.