Hamzeh Ghasemzadeh

Hamzeh Ghasemzadeh, Robert E. Hillman, Daryush D. Mehta

This study's first purpose is to provide quantitative evidence that would incentivize researchers to instead use the more robust method of nested cross-validation. The second purpose is to present methods and MATLAB codes for doing power analysis for ML-based analysis during the design of a study. Monte Carlo simulations were used to quantify the interactions between the employed cross-validation method, the discriminative power of features, the dimensionality of the feature space, and the dimensionality of the model. Four different cross-validations (single holdout, 10-fold, train-validation-test, and nested 10-fold) were compared based on the statistical power and statistical confidence of the ML models. Distributions of the null and alternative hypotheses were used to determine the minimum required sample size for obtaining a statistically significant outcome (α=0.05, 1-\b{eta}=0.8). Statistical confidence of the model was defined as the probability of correct features being selected and hence being included in the final model. Our analysis showed that the model generated based on the single holdout method had very low statistical power and statistical confidence and that it significantly overestimated the accuracy. Conversely, the nested 10-fold cross-validation resulted in the highest statistical confidence and the highest statistical power, while providing an unbiased estimate of the accuracy. The required sample size with a single holdout could be 50% higher than what would be needed if nested cross-validation were used. Confidence in the model based on nested cross-validation was as much as four times higher than the confidence in the single holdout-based model. A computational model, MATLAB codes, and lookup tables are provided to assist researchers with estimating the sample size during the design of their future studies.

Katie Matton, Purvaja Balaji, Hamzeh Ghasemzadeh et al.

Phonotrauma refers to vocal fold tissue damage resulting from exposure to forces during voicing. It occurs on a continuum from mild to severe, and treatment options can vary based on severity. Assessment of severity involves a clinician's expert judgment, which is costly and can vary widely in reliability. In this work, we present the first method for automatically classifying phonotrauma severity from vocal fold images. To account for the ordinal nature of the labels, we adopt a widely used ordinal regression framework. To account for label uncertainty, we propose a novel modification to ordinal regression loss functions that enables them to operate on soft labels reflecting annotator rating distributions. Our proposed soft ordinal regression method achieves predictive performance approaching that of clinical experts, while producing well-calibrated uncertainty estimates. By providing an automated tool for phonotrauma severity assessment, our work can enable large-scale studies of phonotrauma, ultimately leading to improved clinical understanding and patient care.

Hamzeh Ghasemzadeh, Mohammad H. Kayvanrad

Calibration is a common practice in image steganalysis for extracting prominent features. Based on the idea of reembedding, a new set of calibrated features for audio steganalysis applications are proposed. These features are extracted from a model that has maximum deviation from human auditory system and had been specifically designed for audio steganalysis. Ability of the proposed system is tested extensively. Simulations demonstrate that the proposed method can accurately detect the presence of hidden messages even in very low embedding rates. Proposed method achieves an accuracy of 99.3% (StegHide@0.76% BPB) which is 9.5% higher than the previous R-MFCC based steganalysis method.

Hamzeh Ghasemzadeh, Ali Payandeh, Mohammad Reza Aref

Due to hostile environment and wireless communication channel, security mechanisms are essential for wireless sensor networks (WSNs). Existence of a pair of shared key is a prerequisite for many of these security mechanisms; a task that key management system addresses. Recently, an energy efficient method based on public key cryptography (PKC) was proposed. We analyze this protocol and show that it is vulnerable to denial of service (DOS) attacks and adversary can exhaust memory and battery of nodes. Then, we analyze this protocol and show that using a more knowledgeable BS this vulnerability can be solved very efficiently. Based on this observation we propose a modified version of the protocol that achieves immediate authentication and can prevent DOS attacks. We show that the improved protocol achieves immediate authentication at the expense of 1.82 mj extra energy consumption while retaining other desirable characteristics of the basic method.

Hamzeh Ghasemzadeh

Mp3 is a very popular audio format and hence it can be a good host for carrying hidden messages. Therefore, different steganography methods have been proposed for mp3 hosts. But, current literature has only focused on steganalysis of mp3stego. In this paper we mention some of the limitations of mp3stego and argue that UnderMp3Cover (Ump3c) does not have those limitations. Ump3c makes subtle changes only to the global gain of bitstream and keeps the rest of bitstream intact. Therefore, its detection is much harder than mp3stego. To address this, joint distributions between global gain and other fields of mp3 bit stream are used. The changes are detected by measuring the mutual information from those joint distributions. Furthermore, we show that different mp3 encoders have dissimilar performances. Consequently, a novel multi-layer architecture for steganalysis of Ump3c is proposed. In this manner, the first layer detects the encoder and the second layer performs the steganalysis job. One of advantages of this architecture is that feature extraction and feature selection can be optimized for each encoder separately. We show this multi-layer architecture outperforms the conventional single-layer methods. Comparing results of the proposed method with other works shows an improvement of 20.4% in the accuracy of steganalysis.

Hamzeh Ghasemzadeh

Comparing popularity of mp3 and wave with the amount of works published on each of them shows mp3 steganalysis has not found adequate attention. Furthermore, investigating existing works on mp3 steganalysis shows that a major factor has been overlooked. Experimenting with different mp3 encoders shows there are subtle differences in their outputs. This shows that mp3 standard has been implemented in dissimilar fashions, which in turn could degrade performance of steganalysis if it is not addressed properly. Additionally, calibration is a powerful technique which has not found its true potential for mp3 steganalysis. This paper tries to fill these gaps. First, we present our analysis on different encoders and show they can be classified quite accurately with only four features. Then, we propose a new set of calibrated features based on quantization step. To that end, we show quantization step is a band limited signal and steganography noise affects its high frequency components more prominently. By applying a low pass filter on quantization steps, we arrive at an estimation of quantization step, which in turn is used for calibrating the features.

Hamzeh Ghasemzadeh, Meisam Khalil Arjmandi

Calibration and higher order statistics (HOS) are standard components of many image steganalysis systems. These techniques have not yet found adequate attention in audio steganalysis context. Specifically, most of current works are either non-calibrated or only based on noise removal approach. This paper aims to fill these gaps by proposing a new set of calibrated features based on re-embedding technique. Additionally, we show that least significant bit (LSB) is the most sensitive bit-plane to data hiding algorithms and therefore it can be employed as a universal embedding method. Furthermore, the proposed features are based on a model that has the maximum deviation from human auditory system (HAS), and therefore are more suitable for the purpose of steganalysis. Performance of the proposed method is evaluated on a wide range of data hiding algorithms in both targeted and universal paradigms. Simulation results show that the proposed method can detect the finest traces of data hiding algorithms and in very low embedding rates. The system detects steghide at capacity of 0.06 bit per symbol (BPS) with sensitivity of 98.6% (music) and 78.5% (speech). These figures are respectively 7.1% and 27.5% higher than state-of-the-art results based on RMFCC.

Hamzeh Ghasemzadeh, Mohammad H. Kayvanrad

Recently, merging signal processing techniques with information security services has found a lot of attention. Steganography and steganalysis are among those trends. Like their counterparts in cryptology, steganography and steganalysis are in a constant battle. Steganography methods try to hide the presence of covert messages in innocuous-looking data, whereas steganalysis methods try to reveal existence of such messages and to break steganography methods. The stream nature of audio signals, their popularity, and their wide spread usage make them very suitable media for steganography. This has led to a very rich literature on both steganography and steganalysis of audio signals. This paper intends to conduct a comprehensive review of audio steganalysis methods aggregated over near fifteen years. Furthermore, we implement some of the most recent audio steganalysis methods and conduct a comparative analysis on their performances. Finally, the paper provides some possible directions for future researches on audio steganalysis.

Hamzeh Ghasemzadeh, Ali Payandeh, Mohammad Reza Aref

Security is a critical and vital task in wireless sensor networks, therefore different key management systems have been proposed, many of which are based on symmetric cryptography. Such systems are very energy efficient, but they lack some other desirable characteristics. On the other hand, systems based on public key cryptography have those desirable characteristics, but they consume more energy. Recently based on authenticated messages from base station a new PKC based key agreement protocol was proposed. We show this method is susceptible to a form of denial of service attack where resources of the network can be exhausted with bogus messages. Then, we propose two different improvements to solve this vulnerability. Simulation results show that these new protocols retain desirable characteristics of the basic method and solve its deficiencies.

Hamzeh Ghasemzadeh, Mehdi Tajik Khass, Hamed Mehrara

Recently it was shown that permutation-only multimedia ciphers can completely be broken in a chosen-plaintext scenario. Apparently, chosen-plaintext scenario models a very resourceful adversary and does not hold in many practical situations. To show that these ciphers are totally broken, we propose a cipher-text only attack on these ciphers. To that end, we investigate speech permutation-only ciphers and show that inherent redundancies of speech signal can pave the path for a successful cipher-text only attack. For this task different concepts and techniques are merged together. First, Short Time Fourier Transform (STFT) is employed to extract regularities of audio signal in both time and frequency. Then, it is shown that cipher-texts can be considered as a set of scrambled puzzles. Then different techniques such as estimation, image processing, branch and bound, and graph theory are fused together to create and solve these puzzles. After extracting the keys from the solved puzzles, they are applied on the scrambled signal. Conducted tests show that the proposed method achieves objective and subjective intelligibility of 87.8% and 92.9%. These scores are 50.9% and 34.6% higher than scores of previous method.