loading page

Perceptual Hashing of Deep Convolutional Neural Networks for Model Copy Detection
  • +5
  • Chen Haozhe ,
  • Hang Zhou ,
  • Jie Zhang ,
  • Dongdong Chen ,
  • Weiming Zhang ,
  • Kejiang Chen ,
  • Gang Hua ,
  • Nenghai Yu
Chen Haozhe
the CAS Key Laboratory of Electromagnetic Space Information, the CAS Key Laboratory of Electromagnetic Space Information

Corresponding Author:[email protected]

Author Profile
Hang Zhou
Author Profile
Jie Zhang
Author Profile
Dongdong Chen
Author Profile
Weiming Zhang
Author Profile
Kejiang Chen
Author Profile
Nenghai Yu
Author Profile

Abstract

This paper has been accepted by ACM TOMM. https://dl.acm.org/doi/pdf/10.1145/3572777
In recent years, many model intellectual property (IP) proof methods for IP protection have been proposed, such as model watermarking and model fingerprinting. However, as an important part of the model IP protection system, the model copy detection task has not received enough attention. With the increasing number of neural network models transmitted and deployed on the Internet, the search for similar models is in great demand, which concurrently triggers the security problem of copy detection of models for IP protection. Due to the high computational complexity, both model watermarking and model fingerprinting lack the capability to efficiently find suspected infringing models among tens of millions of models. In this paper, inspired by the hash-based image retrieval methods, we propose a perceptual hashing algorithm for convolutional neural networks (CNNs). The proposed perceptual hashing algorithm can convert the weights of CNNs to fixed-length binary hash codes so that the lightly modified version has the similar hash code as the original model. By comparing the similarity of a pair of hash codes between a query model and a test model in the model library, the similar versions of a query model can be retrieved efficiently. To the best of our knowledge, this is the first perceptual hashing algorithm for CNNs. The experiment performed on a model library containing 3,565 models indicates that our proposed perceptual hashing scheme has a superior copy detection performance.