Steganography based on utilizing more surrounding pixels

Abstract

The purpose of this study is to investigate the application of least significant bit insertion method (LSB) in concealing a definite amount of message bits inside a typical media such as an 8-bit gray-scale image and develop an alternative model structure selection algorithm based on considering more surrounding pixels in order to compute the best capacity value for each target pixel. Conventional LSB method’s concept is used as the benchmark for the proposed algorithms. A model structure selection based on not choosing selected pixels as immediate neighbors of the target pixel is proposed in this study to reduce image’s distortion made by embedding random bits or in other words to get a higher imperceptibility. The imperceptibility measures used for evaluation are called peak signal to noise ratio (PSNR), Watson metric and Universal quality index (Q). Previous algorithms utilize at most four (4) out of eight (8) surrounding pixels as multiple based notational system (MBNS) uses three (3) pixels. The effect of considering the rest of surrounding pixels on the performance of the developed model is studied and the effectiveness and shortcomings are highlighted. Results are compared between MBNS, Adaptive and Fixed methods. It is discovered that with similar number of payloads, in most cases, the proposed methods perform better than MBNS in terms of exploring better imperceptibility in predicting more accurate capacity of each target pixel. In addition, the use of pseudo random numbers generator (PRNG) considered for finding the walk path through the image using a stego key and a seed number to provide needed security. Furthermore, the method is proved to be robust against one of the common attacks to stego-images known as Chi-squared attack.