Deep Hashing Based Model for Image Similarity Retrieval in DNA Storage

Abstract

With the rapid development of information technology, the total amount of data generated by human society is growing at an incredible rate, and will soon exceed the storage capacity of existing storage media. DNA molecules, as a natural information storage media, have a broad application prospect because of its high theoretical storage density. With the increase of DNA storage capacity, image similarity retrieval is another important image retrieval method besides random access. DNA hybridization is sequence pairing reaction based on DNA sequence similarity, enabling similarity retrieval in DNA storage. In the context of image similarity retrieval in DNA storage, the key problem is how to map images to DNA sequences while maintaining similarity. In this paper, we model this problem as a variant of deep hashing and then propose a novel DNA-encoded image deep hashing model. We test our model on CIFAR-10 dataset and the results show that our performance is comparable to the state-of-the-art binary-encoded image hashing model, proving that DNA-encoded image similarity retrieval is sufficiently efficient. Our work also provides a novel modeling idea and appropriate evaluation metrics for future research.