Code Verification Hashing for Image Retrieval
Yinqi Chen, Zhiyi Lu, Ya Lu, Yangting Zheng, Peiwen Li, Shuo Kang · 2023
Existing image retrieval deep hashing methods employ full connection as the hash encoding layer and output each bit of hash code in parallel. They regards the hash code as the information of the image, ignoring the relevance between each bit and the redundancy of the whole hash code, resulting in limited performance. Hence, based on the concept of code verification in digital communication (called code verification), the code verification hashing (CVH) is proposed. Different from the exiting methods, CVH constructs hash encoding layer to output the hash code serially, and each output of partial hash code relies on the output of the previous one, so as to make full use of the relevance and redundancy of hash code to generate a more discriminative one. Moreover, CVH also progresses the hash center loss to constraint by all hash centers and makes further improvement. As experiment results, CVH demonstrates excellent retrieval ability.