24-09-2016, 02:25 PM
Index.docx (Size: 664.84 KB / Downloads: 6)
1. Introduction
Signal processing in encrypted domain (SPED) has attracted much attention in recent years. The need for SPED technologies originates from a growing social awareness and relevance of security and privacy. For example, people are increasingly sharing diversity of personal data on the Internet, or doing many works over cloud computing or delegated calculation. SPED was used to accomplish signal processing at the potentially un trusted sites in a privacy-protected form, without or minimally leaking information. Some works have been done on SPED, such as differential-privacy based sanitizing, the buyer–seller watermarking protocol, compression of the encrypted images or videos, reversible data-hiding in encrypted images, and so on. While most SPED algorithms are useful for never-compressed images, few SPED methods are suitable for JPEG, which is the most widely used format.
In some applications, the service provider hopes to reduce the data amount during the transmission, or tends to supply the recipient an encrypted image with a lower resolution. For example, in a dealing system, the seller sends an encrypted image to the server platform, and the server provides the encrypted image with a lower resolution to the buyer prior to the sale completed, during which the image content is not revealed to the server. To the best of authors knowledge, there is no SPED protocol capable of rescaling the encrypted JPEG images for privacy-preserving. In this paper aims to provide a SPED protocol of encrypting the JPEG formatted image, suitable for image rescaling in the encrypted domain.
The protocol is made up of three phases:
1. JPEG image encryption
2. Rescaling the encrypted JPEG image
3. Image decryption.
The image owner generates an encrypted JPEG file by constructing a cipher mask and encrypting the quantized DCT coefficients. After receiving the encrypted JPEG file, the service provider rescales the encrypted image by down sampling the coefficients. On the recipient side, with the encryption keys, the cipher mask is reconstructed and down-sampled as an aid to decrypt the encrypted JPEG file rescaled by the service provider.
Many tools and application based on JPEG encryption, JPEG encryption for image rescaling, compression techniques are available and existed in market. Many tools are available for JPEG encryption in online and offline both mode.
Few tools are listed below :
1) JPEG Crops for Windows (version 0.6.5 beta)
2) Easy Thumbnails from Fookes Software (Windows)
3) PIX resizer from Bluefive Software (Windows)
4) PIX resizer - Free Image Resizer
5) WinZip – Zip UnZip Tool
Literature survey
In recent years, Signal processing in encrypted domain has become interesting research area. A SPED technology originates from a growing social awareness and relevance of security and privacy. For example, people are increasingly sharing diversity of personal data on the Internet, or doing many works over cloud computing or delegated calculation. SPED was used to accomplish signal processing at the potentially un trusted sites in a privacy-protected form, without or minimally leaking information.[1]
1. Some works have been done on SPED, such as
- Differential-privacy based sanitizing
- The buyer–seller watermarking protocol
- Compression of the encrypted images or videos
- Reversible data-hiding in encrypted images and so on.
While most SPED algorithms are useful for never-compressed images, few SPED methods are suitable for JPEG, which is the most widely used format. Aims to provide a SPED protocol of encrypting the JPEG formatted image, suitable for image rescaling in the encrypted domain. The proposed protocol is made up of three phases:
1) JPEG image encryption,
2) Rescaling the encrypted JPEG image
3) Image decryption.
Before sending a JPEG image to the remote server, the original image is encrypted to a meaningless image. On the server side, the service provider rescales the encrypted JPEG image. On the recipient side, the encrypted image is decrypted to a plaintext image with a lower resolution.
1. A sketch of the protocol includes three phases:
1) The image owner
2) The service provider
3) The recipient
The image owner further divides the perturbed version of the quantized DCT coefficients in C’ into MN/256 non-overlapped blocks, each of which is sized 16 x 16. With another key K3, he pseudo randomly permutes the MN/256 blocks to an encrypted version E. This way, structure of the original image is randomized. It should be noted that when the image size is not an integral multiple of 16, we have to pad some zero outside the right and/or bottom boundaries for encoding, which will be removed on the recipient side. Next, by entropy encoding, the image owner compresses the encrypted DCT coefficients in E into another JPEG file J’ . The encrypted file will be sent to the service provider, and the encryption keys and the parameters (K1,K2,K3,T,R) are kept by the owner.
3.2 Rescaling the encrypted JPEG image
With the encrypted JPEG image J’ , although the service provider does not know the content of the image, he can still rescale the encrypted image. Since the encrypted image is in the JPEG format, the service provider first decodes the image J’ into blocks of the quantized coefficients QCr,s(i, j) by entropy decoding, where r = 1,2,. . .,M/8, s = 1,2,. . .,N/8, 1 <= i <= 8 and 1 <= j <= 8. Then he extracts from the file header of J’ the quantization table QT that comprises an 8 x 8 quantization table. Do the inverse quantization of each block with QT to reconstruct the DCT coefficients COr,s(i, j),
〖CO〗_(r,s) (i,j)= 〖QC〗_(r,s) (i,j).QT(i,j) ……………(4)
where r = 1,2, . . .,M/8, s = 1,2, . . .,N/8, 1 <= i <= 8 and 1 <= j <= 8.
The decoded MN coefficients in CO in the encrypted form can be merged into MN/4 coefficients, meaning that the image size is reduced to 1/4 of the original. The algorithm in this paper is adopt to merge every four adjacent coefficient blocks to one coefficient block, in which the adjacent blocks X1~X4 are down-sampled to one 8 x 8 coefficient block X using eq(5)