High-capacity reversible data hiding in encrypted images based on extended run-length coding and block-based MSB plane rearrangement

https://doi.org/10.1016/j.jvcir.2018.12.023Get rights and content

Highlights

  • A high-capacity separable reversible data hiding scheme is proposed.

  • A joint lossless compression scheme is proposed to reserve embedding room.

  • Image recovery and secret data extraction are error-free.

  • The visual quality of the decrypted images is high.

Abstract

In this paper, we propose a novel reversible data hiding method in encrypted images. The proposed method takes full advantage of the spatial correlation in the original images to vacate room for embedding data before image encryption. By jointly using an extended run-length coding and a block-based most significant bit (MSB) plane rearrangement mechanism, the MSB planes of images can be compressed efficiently to generate room for high-capacity embedding. The receiver can extract data directly from encrypted images with only the data hiding key, and the original image or the high-quality plain image that contains secret data can be recovered with only the encryption key. The experimental results prove that the proposed method can reach a high embedding rate and a high PSNR.

Introduction

Reversible data hiding (RDH) is a technique that imperceptibly embeds secret data into the cover data, such as images, audios, videos, and texts, while the embedded data can be extracted and the cover data can be recovered by receivers [1]. The RDH technology is quite useful for some special applications in which images are not allowed to be disturbed, such as military, medical, and law forensics applications. To date, many RDH schemes have been proposed to embed secret data into digital images. These RDH schemes are based on include histogram shifting [2], [3], [4], difference expansion [5], [6], [7], and pixel value ordering [8], [9], [10]. Basically, these RDH schemes in plain images make use of the redundancy information to embed secret data in a reversible manner. If an image has been encrypted, the noise-like encrypted image loses a lot of the redundancy information. Therefore, these methods are not suitable for encrypted images.

With the development of cloud storing and computing, a large number of privacy data, such as videos and images, are stored and processed on cloud servers instead of clients’ terminals. To protect privacy contents, users’ data should be encrypted before being transmitted to cloud servers. To help the data management of encrypted images, RDH in encrypted images (RDHEI) has attracted the attention of many researchers. The RDHEI technology allows the data hider to embed data into encrypted images without knowing the content of the images. Subsequently, receivers can extract embedded data and recover the original image.

The first RDHEI method was proposed in [11]. Recently, Several RDHEI methods have been developed. Some of the methods embed secret data by modifying the pixels of encrypted images and use the spatial correlation of plaintext images to extract the embedded data after image decryption. Zhang [12] proposed a method to embed data by flipping three least significant bits (LSBs) of pixels in encrypted images. First, the encrypted image was divided into blocks of the same size, and each block was divided into two pixel-groups H0 and H1. To embed a bit of 0 or 1 into a block, three LSBs of all pixels in H0 or H1 of the block were flipped. After image decryption, for each block, the receiver flipped three LSBs of all pixels in H0 and H1 respectively to generate two blocks. The original block and the embedded bit can be identified between the two blocks by a smoothness estimation function. Based on [12], several improved RDHEI methods were proposed in [13], [14], [15]. In these studies, more precise functions were designed to estimate the smoothness of the decrypted images, thereby improving the accuracy of data extraction. Liao et al. [14] extended the smoothness function in [12] by taking the pixels in the borders of blocks into account. For each block to be recovered, if any recovered neighboring block exists, the pixels in the adjacent border of the recovered block should be taken into account.

In the previous methods, data extraction and image recovery should be performed jointly. To separate the processes of data extraction and image recovery, separate RDHEI methods have been proposed. Zhang [16] proposed a separable RDHEI method based on LSB compression. The encrypted image was divided into pixel-groups. For each pixel-groups, the LSBs of all pixels were extracted and compressed to reserve embedding room. The secret data can be extracted directly from the embedding room, and image recovery can be performed without data extraction. In [17], half the fourth LSBs of encrypted pixels were compressed by low-density parity-check code (LDPC code) to provide room for embedding data. In [18], Slepian-Wolf source encoding was used to compress the MSBs of the encrypted image to reserve embedding space. In [19], the encrypted image was divided into blocks corresponding to original image’s smooth and complex regions, and the LSBs of the blocks corresponding to smooth regions are compressed for embedding room. In [20], the blocks of the encrypted image were categorized into two sets according to their compressibility to run-length coding. The blocks of two sets are compressed respectively by run-length coding and matrix compression. In [21], the secret data were embedded into the pixels of the encrypted image by MSB substitution. The receiver can extract the secret data from the MSB plane of the encrypted image and recover the image by MSB prediction after decryption. In [22], a block-level predictor was adopted to generate prediction-error sequences from the encrypted image. Based on the prediction-error sequences, a difference expansion scheme was used to embed secret data. In [23], images were encrypted by public key cryptosystems with probabilistic and homomorphism properties, and data were embedded by multi-layer, wet-paper coding. In [24], before encryption, the image was divided into groups, and the groups were encrypted by RC4 with the same key. Since the encrypted groups maintain structure redundancy, a difference histogram is generated according to the encrypted image, and a histogram shifting method is performed for data hiding.

All of the published methods mentioned above involved vacating embedding room after image encryption. Since encryption would minimize the redundancy of images, it is difficult to achieve a satisfactory capacity of embedding space in encrypted images. To improve the embedding rate, some methods have been proposed that reserve room before image encryption (RRBE) [25], [26], [27]. In [25], to vacate embedding room in the LSB planes of the original image, a traditional RDH method was used to embed the LSBs of the selected pixels into other pixels. In [26], the original image was divided into patches, and each patch was compressed by using a sparse coding technology and an over-complete dictionary. In [27], the method divided the MSB planes of the original image into blocks of the same size. Each block was processed respectively according to its distribution of bits. If all bits in a block were identical, the block can be represented by only two bits. If only a few bits in a block were different, the block can be represented as the block type, the number of minority bits, and their locations by less bits. The spared room in the compressed blocks is used to accommodate the original LSBs. After encryption, the data hider can exploit the LSBs of the encrypted image to embed data.

For a plaintext image, spatial correlation exists in neighboring pixels. The neighboring bits in the MSB planes of the image are likely to be identical. Therefore, MSB planes can be transformed into bit streams containing long sequences of the same bits. Motivated from this, we have designed a run-length coding-based compression scheme to heavily compress the MSB planes of the original image for a large embedding capacity.

In this paper, we propose a novel RRBE separable RDHEI method with high capacity. First, we design a joint room vacating technique for plain images, which is composed of extended run-length coding and the block-based MSB plane rearrangement (BMPR) scheme. The joint technique can compress the MSB planes of images efficiently to achieve a lot of spare room. Based on the joint technique, an RDHEI method is proposed. First, the proposed method uses the joint scheme to perform lossless compression on the MSB planes of the plain image to make room, and, then, it embeds LSBs to the compressed MSB planes before encryption. The processed image is encrypted by using the bitwise XOR operation, and the LSBs of the image are reserved for the data hider. The proposed method allows the receiver to use different keys to extract the secret data, decrypt the image, and recover the original image from the embedded encrypted image.

The contributions of this paper are summarized as follows.

  • (1)

    We design a block-based MSB plane rearrangement (BMPR) scheme which transforms the MSB planes of the original image into high compressible bit streams, and an extended run-length coding which compresses the transformed bit streams at a high compression ratio.

  • (2)

    Based on MSB plane compression, we present an RRBE RDHEI method. The method can achieve high embedding capacity and high quality of the marked decrypted image. Data extraction and image recovery of the method are separable and error-free.

The rest of the paper is organized as follows. Section 2 introduces our run-length coding scheme and block-based MSB plane rearrangement scheme. The proposed RDHEI scheme is discussed in detail in Section 3. Section 4 provides the experimental results and comparisons. The conclusions are given in Section 5.

Section snippets

The joint embedding-room-vacating scheme

In this section, we propose a new joint scheme to vacate embedding room in the MSB planes of images. The joint scheme is composed of an extended run-length coding scheme and a block-based MSB plane rearrangement scheme. The basic idea of the joint scheme is to rearrange bits in the MSB planes to construct bit streams that have consecutive long sequences of (0)2 or (1)2 and compress them by extended run-length coding.

Any pixel in the images is constructed of eight bits. Thus, hereafter, the most

The proposed RDHEI method

Fig. 5 shows an overview of the proposed RDHEI method. First, the content owner vacates room for embedding data in the original image. Then, the owner encrypts and shuffles the vacated image using the encryption key and the shuffle key. In the data-embedding phase, the data hider encrypts the secret data using a data hiding key. Then, the data hider reshuffles the encrypted image and embeds the secret data in its LSBs and shuffles the embedded image again with the same shuffle key. The receiver

Experimental results

In this section, we conduct experiments to evaluate the performance of the proposed method, and perform comparisons between the proposed method and several previous RDHEI methods.

First, we evaluate the embedding rates of the proposed method with different block sizes (the length of fixed-length codewords is set to 9), and results are compared with two competitors [24], [27]. The experiments are performed on 1000 512×512×8 grayscale images randomly selected from BOSSBase [29].

Table 3 shows the

Conclusions

In this paper, we propose a new high-capacity reversible data hiding method. The method adopts an extended run-length coding scheme and a block-based MSB plane rearrangement scheme to efficiently compress the MSB planes of the plain images before image encryption, so a large embedding room can be achieved. The hidden data can be extracted completely without error by using only the data hiding key. The original image can be recovered correctly from the decrypted image by using only the

Acknowledgement

This paper is supported by the Natural Science Foundation of Fujian Province, China (2017J05104), and the National Natural Science Foundation of China (61701191).

References (30)

  • M. Li et al.

    Reversible data hiding in encrypted images using cross division and additive homomorphism

    Sign. Process. Image Commun.

    (2015)
  • S. Yi et al.

    Binary-block embedding for reversible data hiding in encrypted images

    Sign. Process.

    (2017)
  • Y.Q. Shi et al.

    Lossless data hiding: fundamentals, algorithms and applications

  • Z. Ni et al.

    Reversible data hiding

    IEEE Trans. Circ. Syst. Video Technol.

    (2006)
  • G. Coatrieux et al.

    Reversible watermarking based on invariant image classification and dynamical error histogram shifting

    IEEE Trans. Inf. Forensics Secur.

    (2013)
  • Cited by (0)

    This paper has been recommended for acceptance by Zicheng Liu.

    View full text