Preprocessing Cover Images for More Secure LSB Steganography

Abstract

We propose a simple method for preprocessing cover images to increase the reliability of random LSB replacement steganography in digital images in spatial domain. After the proposed transformation of cover images, most reliable ste- ganalysis methods in the literature such as RS steganalysis, Sample pair method, Least Square method estimate embedding ratio as almost 100% for any amount of hidden data. Thus it causes inaccurate estimation results by the most reliable LSB steganalysis methods thereby increasing the reliability of LSB replacement steganography. The transformation is based on the count of different types of sample pairs in the image and their subsequent change with embedding. I. I NTRODUCTION Steganography hides the secret message in cover objects to obtain stego objects. Digital images, videos, sound files an d other computer files that contain perceptually irrelevant o r redundant information are used as cover objects to hide secret messages. The goal of steganalysis is to detect/ estimate /retrieve potentially hidden information from observed data with little or no knowledge about the steganographic algorithm or its parameters. The purpose of steganography is to hide the presence of communication, as opposed to cryptography, which aims to make communication unintelligible to those who do not possess the right keys (4). In this paper we concentrate on LSB steganography on digital images stored in uncompressed raw format. Many methods(1), (2), (3), (5), (6), (7), (8), (12) have been prop osed in the literature for steganalysis of digital images. These methods give very accurate results on most of the images. All steganalysis methods are subject to error. Attempts were made in estimating the error in various steganalysis methods. Ker derived error distribution in Least Square steganalysis(9) on images with zero payload. It shows that LSM, one of the most reliable steganalysis methods, gives very accurate results on most of the images, but shows large estimation errors on some images due to image specific properties. Based on the theoretical error model, Ker suggested improvements (12) to the steganalysis method by reducing bias and variance in the case of moderate payloads. However analysis of error distribution in the case of general payloa ds is technically difficult and hence in such cases improvement s suggested can not be used for estimation (12). With the development of very accurate steganalysis techniques, methods have been proposed for increasing reliability of LSB steganography also. Fridrich et. al deve loped a general coding method called matrix embedding(10) that can be applied to most steganographic schemes for improving their steganographic security. Luo et.al (11) developed a method for increasing security of LSB steganography based on chaos system and dynamic compensation. The dynamic compensation is done after hiding the data in the image. After doing dynamic compensation the most accurate methods like RS, SPM and LSM and their improved versions detect stego images with very high payload as cover images. However dynamic compensation causes the cover images to be detected as cover images only. In this paper we present a simple transformation on cover images to cause larger estimation errors for the most reliab le steganalysis techniques such as RS steganalysis, Least Square method and Sample pair method. The method we propose involves a very simple operation of flipping for preprocessi ng a cover image in such a way that it gives inaccurate estimation results with/ without hidden data.