CONTAINER`S STATISTIC FEATURES CHANGES FOR STEGANOGRAPHIC ALGORITHM IN FREQUENCY DOMAIN
Abstract
The subject of this work done is digital steganography field, namely hidden messages detec-tion in statical digital pictures problem. In this paper probability of hidden messages detection using statisti-cal features distribution of discrete cosine transform coefficients for two steganographic algorithms was checked. This data allow detecting messages hidden by steganographic algorithm with DCT coefficient re-placement and improving algorithm with image statistic features considering. For this purpose was used early proposed parameter: imbalance of container`s paired discrete cosine transform coefficients. As ste-ganographic containers, JPEG images were considered. For image DCT coefficients quantization, standard IJG matrix was used. These parameters were measured for one thousand of empty containers samples and for equal quantity of containers with messages, hidden by algorithm with DCT coefficients replacement and by early proposed steganographic algorithm with container image statistics accounting. For all ste-ganographic containers same text message was embedded that container`s statistic changes depend only on its own characteristics. Imbalance of container`s paired discrete cosine transform coefficients was used as a measured parameter for embedded message detection. Based on the obtained experimental results, the initial and alternative hypotheses about message presence in container were considered. Also measured parameter threshold was chosen on condition, that probability of type I and type II errors is equal. With this parameter threshold value probability of type I and type II errors, probability of hidden messages detection and statisti-cal significance of chosen parameter α was calculated. Based on the received results, only messages hidden by steganographic algorithm with DCT coefficient replacement can be definitely detected by chosen parame-ter measurement. For embedding with image statistic features considering probability of message detecting nearly equal to random guessing results.
