Content-adaptive reversible visible watermarking in encrypted images

Highlights

• Data embedding positions for accommodating the watermark can be adaptively selected to achieve the tradeoff between watermark visibility and marked image quality.

• A novel framework for reversible visible watermarking in encrypted images is presented.

• The original cover image can be losslessly recovered after image decryption and watermark extraction.

Abstract

The reversible visible watermark which serves as the perceptual ownership identifier can be extracted to losslessly recover the original cover media. This paper presents a novel content-adaptive reversible visible watermarking scheme in encrypted images. To achieve the tradeoff between watermark visibility and marked image quality, data embedding positions for accommodating the watermark are adaptively selected using the visual perceptual model before encryption. Due to weak spatial correlation in encrypted images, the data embedding room is vacated before encryption with a traditional reversible data hiding algorithm to contain pixel bits in data embedding positions. Therefore, it is convenient for the data hider to embed the visible watermark in encrypted images by substituting pixel bits in data embedding positions. If the receiver decrypts the marked encrypted image without extracting the embedded watermark, the visibly marked image can be obtained. In addition, if the receiver decrypts the marked encrypted image and extracts the embedded watermark, the original image can be perfectly recovered. Experimental results demonstrate the merits of the proposed scheme in terms of marked image quality, watermark visibility and watermark robustness.

Introduction

Visible watermarking is a technique which perceptibly embeds a watermark in the cover digital media to identify the ownership and deter malicious attempts of copyright violations. Visible watermarks can be company logos, ownership descriptions, and personal digital signatures, etc. Compared with invisible watermarks, visible watermarks can present ownership information directly on the marked media. The key desirable characteristic of visible watermarking is that the embedded watermark should not significantly obscure the marked media details beneath it [1], [2].

Watermark embedding degrades the cover media quality in general. Reversible data hiding (RDH) serves as a technique which embeds data into the cover digital media so that the embedded data can be extracted to losslessly recover the original cover [3]. Lossless recovery of the original cover is necessary in some application scenarios (e.g., law forensics, historical art imaging, and medical image analysis) where serious concerns about image quality exist. Fortunately, the reversible visible watermarking scheme can be implemented by combining reversible data hiding and visible watermarking.

Many reversible visible image watermarking schemes are proposed in the past [4], [6], [7], [8], [9], [10], [12], [13], [14]. Hu and Jeon [4] proposed a bit plane alteration-based scheme for visible watermark embedding, in which partial one-bit pixels in the watermark embedding region are compressed by the JBIG algorithm [5] and substituted with the to-be-embedded watermark. To achieve the reversibility, additional payload which consists of compressed one-bit pixels should be embedded in the marked image. The watermark embedding capacity is controlled by data compression efficiency in this scheme [4]. Yip et al. [6] presented two reversible visible watermarking algorithms based on pixel value matching and pixel position shift. In [7], [8], reversible visible watermarking schemes using one-to-one mapping of image pixels are elaborated. Yang et al. [9] proposed to reveal the reversible visible watermark through adaptively adjusting the pixel values and embed the reconstruction packet to restore the original cover image. Chen et al. [10] proposed to reversibly embed the visible watermark using the conventional difference expansion technique [11]. Yang et al. [12] improved the scheme in [10] to reduce the number of overflow/underflow marked pixels whose values are larger than 255 or less than 0. In the scheme proposed by Mohammad et al. [13], pixel circular shift operation is conducted to reversibly embed the visible watermark in the block truncation coding-compressed image. The watermark can be extracted according to the parity of the bit plane. Lin et al. [14] achieved the reversible visible watermarking scheme in DCT domain. However, the original cover image can be recovered only in the case that the original watermark image is obtained.

With the prosperity of cloud computing and mobile network, signal processing in the encrypted domain has gained increasing research interests, such as feature extraction [15], image compression [16], and data hiding [17], [18]. Consequently, the research on reversible data hiding in encrypted images driven by the needs from cloud platforms and privacy preservation has attracted considerable attention [19], [20], [21], [22], [23], [24], [25]. Because image authentication [26], [27], [28] serves as the important application scenario of data hiding, the data hider can embed additional authentication data into encrypted images for access control and media annotation without leaking the privacy of the image content owner. Most reversible data hiding schemes in encrypted images aim to invisibly embed additional data. In some application scenarios, reversible visible watermarking in encrypted images which can convey ownership information directly on the marked media is also desirable. However, the open innovation literature has so far witnessed few attempts to explore this subject. In [29], Zhang et al. proposed a reversible visible watermarking scheme in encrypted images using wet paper codes [30]. The exclusive-or operation is applied for image encryption in this scheme. In addition, partial encrypted data corresponding to black pixels of the binary watermark image should be modified to insert the visible watermark and contain some payload for image recovery. The watermark is invisible in the encrypted domain although it is embedded in the encrypted image. After direct decryption operation, the embedded watermark can be visible.

Generally, reversible visible watermarking in encrypted images requires that the embedded watermark is visible yet should not significantly obscure the marked image details beneath it. These two requirements usually conflict with each other. How to obtain the tradeoff between these two conflicting requirements is the key problem of reversible visible watermarking. If the embedded watermark energy is increased to enhance its visibility, image quality degradation will be severer and vice versa. As discussed, many reversible visible image watermarking schemes have been proposed in the past few years. However, the basic issues corresponding to watermark visibility and marked image quality have not been resolved. Hence, new reversible visible watermarking schemes in encrypted images which are capable of achieving the tradeoff between watermark visibility and marked image quality should be sought.

We propose a novel content-adaptive reversible visible watermarking scheme in encrypted images in this paper. In our proposed scheme, we concentrate on adaptively selecting data embedding positions for accommodating the watermark using the visual perceptual model before encryption to achieve the tradeoff between watermark visibility and marked image quality. Considering the weak spatial correlation in encrypted images, the data embedding room is vacated before encryption with a traditional reversible data hiding algorithm to carry pixel bits in data embedding positions. Therefore, the visible watermark can be embedded in encrypted images by substituting pixel bits in data embedding positions.

In conclusion, the highlights of this paper can be summarized as follows.

• Data embedding positions for accommodating the watermark can be adaptively selected to achieve the tradeoff between watermark visibility and marked image quality.

• A novel framework for reversible visible watermarking in encrypted images is presented.

• The original cover image can be losslessly recovered after image decryption and watermark extraction.

The remainder of this paper is organized as follows. The scheme for content-adaptive reversible visible watermarking in encrypted images is elaborated in Section 2 followed by some implementation issues in Section 3. The experimental results and analysis are presented in Section 4. Finally, Section 5 concludes the paper.