| Author: | Wang, Benxuan |
| Title: | Image encryption technique for privacy preserving image retrieval |
| Advisors: | Lo, Kwok-tung (EEE) |
| Degree: | Ph.D. |
| Year: | 2025 |
| Department: | Department of Electrical and Electronic Engineering |
| Pages: | iv, 84 pages : color illustrations |
| Language: | English |
| Abstract: | Recently, along with the rapid development of multimedia techniques, many images have been generated daily over various network platforms and those images can be stored to cloud servers for convenience. However, the rich sensitive information embedded in those images often results in security and privacy issues when personal images are outsourced. So the need for secure storage and transmission of images has become increasingly important. However, image retrieval methods require the image information before it can be retrieved. Therefore, it would be necessary to search for effective image encryption and retrieval techniques to protect privacy and also maintain the availability of encrypted data. To handle the conflict, thesis presents our contributions in realizing the encrypted image retrieval system and three different encrypted image retrieval schemes are proposed. The proposed three models can broadly categorize into pixel based method, DCT coefficient based method and deep network based method. In the first encrypted image retrieval scheme, the encryption operations are adopted on image pixels. For a given image, we first divide it into 8x8 non-overlapped blocks due to the JPEG standard. And the 8-bit binary sequences of pixels in each block are confused by two-level permutation. Specifically, more significant 4-bit binary sequence of the pixel is confused by block permutation, while intra-block permutation is conducted on the less significant 4-bit binary sequence. After encryption on binary sequence, the image confusion is used by block permutation to increase image security and the index is generated from a logistic map. The histogram features can be generated from the confused blocks directly for retrieval processing. In the second encrypted image retrieval scheme, we extract the features for retrieval from the frequency domain which would consume less computation and communication resources. DCT coefficients are utilized to obtain feature vectors. The encryption operations, including coefficient value substitution and intra-block coefficient shuffling, are performed on JPEG images. With the proposed encryption and compression scheme, the feature would be directly extracted from the frequency domain by using the learning network. And the Siamese architecture for metric learning is used for capturing the similarity well. Finally, the user can receive several encrypted images with similar content according to the query. As for the third encrypted image retrieval scheme, a new image compression and encryption framework is proposed which integrates encryption algorithms with a learning-based compression network. Our model employs Auto-Encoder (AE) based compression network as the backbone and encryption layers are added. And for higher security, the parameters of the synthesis network are replaced by a new parameter matrix based on a logistic map controlled by a secret key. The encryption key of the system is derived from the image content, which will be embedded in the deep feature vectors. And to learn the entropy model from the scrambled feature maps, an attention scheme is exploited in estimating parameters to achieve more effective compression. In this scheme, the encrypted feature maps are the inputs of another deep network for retrieval. And the training loss function for this retrieval model consists of ranked list loss and cross-entropy loss. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13955