Post-processing for Enhancing Audio Steganographic Undetectability
Currently, the conventional steganography method often only perform data embedding, without additional post-processing to enhance undetectability. In this work, we propose a new audio post-processing steganography model, which further hiding the traces to
- PDF / 738,163 Bytes
- 14 Pages / 439.37 x 666.142 pts Page_size
- 23 Downloads / 197 Views
Abstract. Currently, the conventional steganography method often only perform data embedding, without additional post-processing to enhance undetectability. In this work, we propose a new audio postprocessing steganography model, which further hiding the traces to a certain extent. Specifically, we design the Signal-to-Noise Ratio (SNR) threshold to determine whether the current stego is suitable for adding disturbance or not, and use JS divergence to decide whether the added disturbance is kept or not, respectively. The designed two measures will process the traces frame-by-frame by adding appropriate disturbances on needed sampling points of the stego audio. Experimental results illustrate that, with the proposed post-processing, the undetectability can be successfully improved without affecting the message extraction. Keywords: Steganographic traces
1
· Post-processing · JS divergence.
Introduction
Digital steganography aims at embedding the secret message into the cover (e.g., audio, image, video, etc.) while attaining the undetectability of the embedded message. In the past decade, for the mainstream steganography schemes, one long-standing crucial problem is how to minimize the distortion during embedding. Designing a distortion cost function to minimize the embedded distortion became the most popular solution, since Filler et al. [3] proposed the seminal Syndrome-Trellis Codes (STCs) framework, in 2011. Inspired by Filler’s work, in the field of image steganography, there were many excellent algorithms under this STC framework, e.g., HUGO [13], WOW[5], UNIWARD [6], HILL [7], and MiDOP [15], to name a few. Those methods sequentially embedded message into low-cost regions of the image and update the cost to minimize it dynamically. In the field of audio steganography, there also were some representative works developed audio steganography schemes under a similar framework. Luo et al. [10] used the residual of the signal before and after Advanced Audio Coding compression to calculate the possible distortion. Gong et al. [4] used the short-term stability of pitch delay and the statistical distribution of adjacent subframe to design a distortion function for the c Springer Nature Singapore Pte Ltd. 2020 S. Yu et al. (Eds.): SPDE 2020, CCIS 1268, pp. 546–559, 2020. https://doi.org/10.1007/978-981-15-9129-7_38
Post-processing for Enhancing Audio Steganographic Undetectability
547
Adaptive Multi-Rate (AMR) speech stream. Yi et al. [17] considered integrating the psychoacoustic model of intra-frame with frame-level perceptual distortion of inter-frame as a distortion function. Chen et al. [2] proposed a rule named “large-amplitude-first,” which would assign low modification distortion to large amplitude audio samples. All the above methods involved the cover selection, which evaluates the possible distortion cost of each pixel or sampling point, and designing a distortion cost function. By adaptively selecting suitable areas for embedding, all methods are able to find the most reasonable embedding strategy with mi
Data Loading...
