SECURITY ENHANCEMENT FOR TPVD STEGANOGRAPHIC METHOD Nazanin Zaker, Ali Hamzeh CSE and IT Dept., ECE School, Shiraz University, Shiraz, Iran ABSTRACT The steganographic method named Tri-way Pixel Value Differencing (TPVD) which is a modified version of original Pixel Value Differencing (PVD) steganography, has significantly improved the embedding capacity of original method by embedding secret bits in all horizontal, vertical and diagonal edges of cover images. However, we have shown in this paper that TPVD is drastically vulnerable to statistical analysis of histogram of pixel differences. The occurrence of unusual steps in the derived histogram of pixel differences demonstrates the existence of hidden message. In order to enhance the security, we applied modifications on TPVD steganography and made it secure against current steganalysis. In the proposed modified algorithm, TPVD’s embedding phase is mixed with our previously proposed method named EPVD, and the visual characteristics of cover image are genuinely maintained. Index Terms— Data Hiding, steganography, steganalysis, Histogram Quantization, PVD, TPVD, EPVD 1. INTRODUCTION Data security has become an important issue in modern world. Wide spread and unsecure network connections have urged people to seeking ways in order to ensure the safety of their data. Steganography which is known as the science of hiding secret messages in another seemingly innocent carrier has received considerable attention recently. In contrast to cryptography that draws the attention of eavesdroppers, the very existence of secret messages is not perceptible in steganography [1]. The digital carrier that used in modern steganography can be audio, video, email, message, disk space or image [2]. Through the vast transmission of images over internet and their considerable capacity for hiding secret data, they have become one of the most important carriers. After embedding secret message in the carrier or cover image, a so called stego-image is achieved [2]. A secret key is used for embedding and it is essential to have this secret key to extract the hidden message [3]. Moreover, the cover image should be kept secret and exact comparison of cover image with stego-image can easily reveal the existence of embedded data. In contrary to steganography, steganalysis try to uncover and detect hidden messages. They identify the weaknesses of steganographic methods and track basic attributes of cover media that have been altered inevitably by steganographic algorithms [4]. Many steganographic techniques have been proposed that use different domains of images for embedding. One of the recently proposed methods is Pixel Value Differencing (PVD) steganography which provides a high capacity for data embedding in spatial domain of images and maintains the visual characteristics of stego-images [5]. However, this steganography can be detected by histogram base analysis of pixel differences [6, 7]. In order to strengthen PVD steganography, many modifications have been proposed that mostly aim to increase the security of hidden data against steganalysis attacks. One of the famous modifications on PVD introduces variable range intervals in order to protect this steganography from histogram base attacks [6]. However, this method has recently been detected by a neural network based steganalysis approach from the pixel difference histogram [8]. An enhancement of PVD, named EPVD, has been recently proposed which improves PVD’s security to a great proportion and cannot be detected via current steganalysis [9]. Tri-way PVD (TPVD) is another new modified version suggested in [10] which increases the embedding capacity to 150 percent and returns stego- images with good visual qualities [10]. To the best of our knowledge, no steganalysis attack has been proposed against TPVD so far. In this paper a steganalysis is proposed that is able to detect TPVD steganography in stego-images. Besides, original TPVD is modified via EPVD steganography and its security has been highly enhanced. In section 2, PVD and TPVD steganographic methods are briefly reviewed. Proposed steganalysis and the modified scheme are explained in section 3. Experimental results and conclusion of the paper are found in sections 4 and 5 respectively. 2. PVD AND TPVD STEGANOGRAPHY In PVD steganography that was proposed in [5], the cover image is a gray valued image. In order to embed secret message, first, the cover image is scanned in a zigzag manner to produce an image vector [7]. Figure 1, shows how the cover image is scanned in PVD method. Second, the image vector is segmented into separable blocks of two consecutive pixels. The difference between pixel values of a block is used for embedding. Let the pixel values in a block be ሺ݌ ଵ ,݌ ଶ ሻ. The difference  is calculated as  ൌ ݌ ଶ –݌ ଵ that is between െʹ55 and ʹ55. The greater values of || indicates the edge areas in an image and its smaller values show the smooth areas of an image. According to human eyes sensitivity, more data can be stored in edge areas of images and less data in smooth areas. In PVD steganography, these difference values are classified into continuous ranges say  ௜ ,  ൌ ͳ, … , . Lower bound and upper bound of these ranges can be shown as  ௜ and ݑ ௜ , so that  ௜ ൌ ሾ ௜ ݑ ௜ ሿ. One applicable range intervals that we used in this paper is R={[0 7], [8 15], [16 31], [32 63], [64 127], [128 255]}. The width of each range is calculated as ݓ ௜ ൌ ݑ ௜ –  ௜ ൅ͳ, and indicates the number of bits that can be hidden in a block. For each block, if  א ௜ , | log ଶ ݓ ௜ | secret bits would be taken and converted into decimal value . The new difference value Ԣ will be computed via the following equation. 206 2010 5th International Symposium on Telecommunications (IST'2010) 978-1-4244-8184-2/10/$26.00 ©2010 IEEE