Jehyun Lee

SRMT: A Lightweight Encryption Scheme for Secure Real-time Multimedia Transmission

Securing multimedia transmission has become a challenging issue due to the popularization of real-time multimedia applications such as video surveillance, satellite communication and web cams. However, previous security algorithms do not always guarantee a satisfactory degree of media quality and latency. In order to provide both security and media QoS, a viable security mechanism must consider three properties: processing time, compression rate and security level. In this paper, we propose a light-weight encryption scheme without loss of security and media QoS, called secure real-time media transmission (SRMT) using two block transpositions and a XOR operation. SRMT is studied with respect to MPEG-4, which is widely used in todays multimedia applications. Experimental results with various MPEG-4 movies show that the SRMT scheme achieves real-time transmission of encrypted media data without loss of security and media QoS. Though SRMT is conducted on uncompressed raw data, SRMT encrypts 3 times faster than the AES encryption of MPEG compressed data. Also, we show that manipulating key frames and a compression method can lessen increasing ratio of encrypted MPEG size, e.g., 70.5% improvement over an existing combination method of block transpositions and XOR operations.

PsyBoG: A scalable botnet detection method for large-scale DNS traffic

Domain Name System (DNS) traffic has become a rich source of information from a security perspective. However, the volume of DNS traffic has been skyrocketing, such that security analyzers experience difficulties in collecting, retrieving, and analyzing the DNS traffic in response to modern Internet threats. More precisely, much of the research relating to DNS has been negatively affected by the dramatic increase in the number of queries and domains. This phenomenon has necessitated a scalable approach, which is not dependent on the volume of DNS traffic. In this paper, we introduce a fast and scalable approach, called PsyBoG, for detecting malicious behavior within large volumes of DNS traffic. PsyBoG leverages a signal processing technique, power spectral density (PSD) analysis, to discover the major frequencies resulting from the periodic DNS queries of botnets. The PSD analysis allows us to detect sophisticated botnets regardless of their evasive techniques, sporadic behavior, and even normal users’ traffic. Furthermore, our method allows us to deal with large-scale DNS data by only utilizing the timing information of query generation regardless of the number of queries and domains. Finally, PsyBoG discovers groups of hosts which show similar patterns of malicious behavior. PsyBoG was evaluated by conducting experiments with two different data sets, namely DNS traces generated by real malware in controlled environments and a large number of real-world DNS traces collected from a recursive DNS server, an authoritative DNS server, and Top-Level Domain (TLD) servers. We utilized the malware traces as the ground truth, and, as a result, PsyBoG performed with a detection accuracy of 95%. By using a large number of DNS traces, we were able to demonstrate the scalability and effectiveness of PsyBoG in terms of practical usage. Finally, PsyBoG detected 23 unknown and 26 known botnet groups with 0.1% false positives.

Tracking multiple C&C botnets by analyzing DNS traffic

Botnets have been considered as a main source of Internet threats. A common feature of recent botnets is the use of one or more C&C servers with multiple domain names for the purpose of increasing flexibility and survivability. In contrast with single domain botnets, these multi domain botnets are hard to be quarantined because they change domain names regularly for connecting their C&C server(s). In this paper, we introduce a tracking method of botnets by analyzing the relationship of domain names in DNS traffic generated from botnets. By examining the DNS queries from the clients which accessed the known malicious domain names, we can find a set of unknown malicious domain names and their relationship. This method enables to track malicious domain names and clients duplicately infected by multiple bot codes which make botnets revivable against existing quarantine methods. From the experiments with one hour DNS traffic in an ISP network, we find tens of botnets, and each botnet has tens of malicious domains. In addition to botnet domains, we find a set of other domain names used for spamming or advertising servers. The proposed method can be used for quarantining recent botnets and for limiting their survivability by tracking the change of domain names.

Hidden bot detection by tracing non-human generated traffic at the Zombie host

Defeating botnet is the key to secure Internet. A lot of cyber attacks are launched by botnets including DDoS, spamming, click frauds and information thefts. Despite of numerous methods have been proposed to detect botnets, botnet detection is still a challenging issue, as adversaries are constantly improving bots to write them stealthier. Existing anomaly-based detection mechanisms, particularly network-based approaches, are not sufficient to defend sophisticated botnets since they are too heavy or generate non-negligible amount of false alarms. As well, tracing attack sources is hardly achieved by existing mechanisms due to the pervasive use of source concealment techniques, such as an IP spoofing and a malicious proxy. In this paper, we propose a host-based mechanism to detect bots at the attack source. We monitor nonhuman generated attack traffics and trace their corresponding processes. The proposed mechanism effectively detects malicious bots irrespective of their structural characteristics. It can protect networks and system resources by shutting down attack traffics at the attack source. We evaluate our mechanism with eight real-life bot codes that have distinctive architectures, protocols and attack modules. In experimental results, our mechanism effectively detects bot processes in around one second after launching flood attacks or sending spam mails, while no false alarm is generated.

Screening smartphone applications using malware family signatures

The sharp increase in smartphone malware has become one of the most serious security problems. Since the Android platform has taken the dominant position in smartphone popularity, the number of Android malware has grown correspondingly and represents critical threat to the smartphone users. This rise in malware is primarily attributable to the occurrence of variants of existing malware. A set of variants stem from one malware can be considered as one malware family, and malware families cover more than half of the Android malware population. A conventional technique for defeating malware is the use of signature matching which is efficient from a time perspective but not very practical because of its lack of robustness against the malware variants. As a counter approach for handling the issue of variants behavior analysis techniques have been proposed but require extensive time and resources. In this paper, we propose an Android malware detection mechanism that uses automated family signature extraction and family signature matching. Key concept of the mechanism is to extract a set of family representative binary patterns from evaluated family members as a signature and to classify each set of variants into a malware family via an estimation of similarity to the signatures. The proposed family signature and detection mechanism offers more flexible variant detection than does the legacy signature matching, which is strictly dependent on the presence of a specific string. Furthermore, compared with the previous behavior analysis techniques considering family detection, the proposed family signature has higher detection accuracy without the need for the significant overhead of data and control flow analysis. Using the proposed signature, we can detect new variants of known malware efficiently and accurately by static matching. We evaluated our mechanism with 5846 real world Android malware samples belonging to 48 families collected in April 2014 at an anti-virus company; experimental results showed that; our mechanism achieved greater than 97% accuracy in detection of variants. We also demonstrated that the mechanism has a linear time complexity with the number of target applications.

Screening Smartphone Applications Using Behavioral Signatures

The sharp increase of smartphone malwares has become one of the most serious security problems. The most significant part of the growth is the variants of existing malwares. A legacy approach for malware, the signature matching, is efficient in temporal dimension, but it is not practical because of its lack of robustness against the variants. A counter approach, the behavior analysis to handle the variant issue, takes too much time and resources. We propose a variant detection mechanism using runtime semantic signature. Our key idea is to reduce the control and data flow analysis overhead by using binary patterns for the control and data flow of critical actions as a signature. The flow information is a significant part of behavior analysis but takes high analysis overhead. In contrast to the previous behavioral signatures, the runtime semantic signature has higher family classification accuracy without the flow analysis overhead, because the binary patterns of flow parts is hardly shared by the out of family members. Using the proposed signature, we detect the new variants of known malwares by static matching efficiently and accurately. We evaluated our mechanism with 1,759 randomly collected real-world Android applications including 79 variants of 4 malware families. As the experimental result, our mechanism showed 99.89% of accuracy on variant detection. We also showed that the mechanism has a linear time complexity as the number of target applications. It is fully practical and advanced performance than the previous works in both of accuracy and efficiency.

DroidGraph: discovering Android malware by analyzing semantic behavior

Mobile malware has been recently recognized as a significant problem in accordance with the rapid growth of the market share for smartphones. Despite of the numerous efforts to thwart the growth of mobile malware, the number of mobile malware is getting increased by evolving themselves. By applying, for example, code obfuscation or junk code insertion, mobile malware is able to manipulate its appearance while maintains the same functionality, thus mobile malware can easily evade the existing anti-mobile-malware solutions. In this paper, we focus on Android malware and propose a new method called DroidGraph to discover the evolved Android malware. DroidGraph leverages the semantics of Android malware. More precisely, we transform an APK file for Android malware to hierarchical behavior graphs that represent with 136 identical nodes based on the semantics of Android API calls. Then, we select unique behavior graphs as semantic signatures describing common behaviors for Android malware. In evaluation, DroidGraph shows approximately 87% of detection accuracy with only 40 semantic signatures against 260 real-world Android malware, and no false positives for 3,623 benign applications.

PsyBoG: Power spectral density analysis for detecting botnet groups

Botnets are widely used for acquiring economic profits, by launching attacks such as distributed denial-of-service (DDoS), identification theft, ad-ware installation, mass spamming, and click frauds. Many approaches have been proposed to detect botnet, which rely on end-host installations or operate on network traffic with deep packet inspection. They have limitations for detecting botnets which use evasion techniques such as packet encryption, fast flux, dynamic DNS and DGA. Sporadic botnet behavior caused by disconnecting the power of system or botnets own nature also brings unignorable false detection. Furthermore, normal users traffic causes a lot of false alarms. In this paper, we propose a novel approach called PsyBoG to detect botnets by capturing periodic activities. PsyBoG leverages signal processing techniques, PSD (Power Spectral Density) analysis, to discover the major frequencies from the periodic DNS queries of botnets. The PSD analysis allows us to detect sophisticated botnets irrespective of their evasion techniques, sporadic behavior and even the noise traffic generated by normal users. To evaluate PsyBoG, we utilize the real-world DNS traces collected from a /16 campus network including more than 48,046K queries, 34K distinct IP addresses and 146K domains. Finally, PsyBoG caught 19 unknown and 6 known botnet groups with 0.1% false positives.