Sequential Picture Encryption Using RNN Prediction for Face Data Security
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing face swiping data encryption methods are vulnerable to cracking, leading to poor safety due to the use of asymmetric algorithms with non-neural network safety factors.
Innovation Solution
Implement a face picture encryption method that involves feature extraction, target prediction, and encryption based on prediction information, using a recurrent network to consider the relevance of picture features in chronological order, enhancing security by avoiding easy cracking.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If asymmetric encryption algorithm with timestamp or counter is used to encrypt face swiping data, then encryption format is established, but security is poor because the encryption format is easily cracked
Solution Approach 1:
The patent applies dynamics by transitioning from static encryption (using fixed timestamp or counter) to dynamic encryption (using recurrent neural network that processes sequential picture features). The RNN updates hidden states dynamically based on previous picture features, creating a time-varying encryption key that adapts to the sequence of face pictures, thereby improving security without excessive complexity
Solution Approach 2:
The patent changes the encryption parameters from simple timestamp/counter to complex prediction information generated by RNN. The encryption key is transformed from static values to dynamic predictions based on sequential picture features, fundamentally changing the parameter space and making the encryption format much harder to crack
2Reliability
If recurrent neural network is used to perform target prediction on picture features, then security is improved by considering temporal relevance, but computational complexity increases
Solution Approach 1:
The patent applies preliminary action by pre-training the recurrent neural network offline before actual encryption operations. The RNN is trained on large datasets to learn temporal patterns in advance, so that during runtime, only inference is needed rather than full training, significantly reducing the computational power required during actual encryption while maintaining high security
Solution Approach 2:
The patent uses copying by training the RNN model once on comprehensive data and then copying this trained model for deployment. Instead of performing complex computations repeatedly, the learned patterns are copied into the model weights, allowing efficient inference with reduced computational power while maintaining the security benefits of temporal analysis
Data Source
Figure 1~2
Figure 3
Figure 4
AI summary
Disclosed are a picture encryption method and apparatus, a computer device, a storage medium and a program product. The method includes: obtaining N first pictures having a chronological order (301); performing feature extraction on the N first pictures to obtain one or more picture features of each of the N first pictures (302); successively performing target prediction on the N first pictures according to the chronological order to obtain prediction information of each of the N first pictures (303), where the target prediction is prediction on the first pictures based on status information, and the status information is information updated based on the one or more picture features of each of the first pictures on which the target prediction has been performed; and encrypting the N first pictures based on the prediction information of each of the N first pictures (304). Through the above-mentioned solution, a condition that a first image is easily cracked for an encryption format only introducing a non-neural network safety factor is avoided, so that the safety of the encrypted first image is also improved while the encryption format that encrypts the first pictures is expanded.