MSU Computer Science and Engineering Technical Reports

Top-Down Connections in Self-Organizing Hebbian Networks: Topographic Class Grouping

2011-05-12

Matching Forensic Sketches to Mugshot Photos

2011-05-12

Authors: Brendan Klare, Zhifeng Li, Anil K. Jain

The problem of matching a forensic sketch to a gallery of mugshot images is addressed in this paper. Previous research in sketch matching offered solutions to matching highly accurate sketches that were drawn while looking at the subject. We refer to these as viewed sketches. Forensic sketches differ from viewed sketches in that they are drawn by a police sketch artist using the description provided by an eye-witness who is typically unfamiliar with the subject. To solve the problem of matching forensic sketches, we propose a robust framework, called local feature-based discriminant analysis (LFDA). In LFDA, we first represent both sketches and photos using gradient location and orientation histograms and multi-scale local binary patterns. A discriminant projection method is then used on the feature-based representation for minimum distance matching. We apply the LFDA method to match a dataset of forensic sketches against a mugshot gallery containing 10,000 images. Compared to a leading commercial face recognition system, LFDA offers substantial improvements to matching forensic sketches to the corresponding images. We are able to further improve the matching performance using race and gender information to reduce the target gallery size. Experiments on matching viewed sketches also demonstrate that LFDA matching method exceeds all other previous methods of sketch matching.

PILL-ID: Matching and Retrieval of Drug Pill Imprint Images

2011-05-12

Authors: Youngbum Lee, Unsang Park, Anil K. Jain

Automatic illicit drug pill matching and retrieval is becoming an important problem due to an increase in the number of tablet type illicit drugs in our society. We propose an automatic method to match drug pill images based on the imprints appearing on the tablet. This will help identify the source and manufacturer of the illicit drugs. The feature vector extracted from tablet images is based on edge localization and invariant moments. Instead of storing a single template for each pill type, we generate multiple templates during the edge detection process. This circumvents the difficulties during matching due to variations in illumination and viewpoint. Experimental results using a set of real drug pill images (822 illicit drug pill images and 1,294 legal drug pill images) showed 77.0% (91.0%) rank-one (rank-20) matching accuracy.

Grammar string: a novel data structure for ncRNA secondary structure modeling

2011-05-12

Authors: Yanni Sun, Seyedeh Shohreh Takyar, Rujira Achawanantakun

Motivation: Noncoding RNA (ncRNA) identification is an important research problem in computational biology. The state-of-the-art method for ncRNA identification is comparative ncRNA analysis, which searches for homologous ncRNAs in multiple genomes. An ncRNA's function is determined by both its sequence and its secondary structure. Thus ncRNA homology search algorithms must detect both sequence and structural similarity. To this end, different secondary structure modeling and comparison methods have been developed. However, many existing tools rely on complicated data structures for structure modeling, incurring high computational cost. A more efficient ncRNA secondary structure modeling and comparison algorithm is thus highly desirable for genome-scale ncRNA homology search. Results: A novel data structure named grammar string is designed to represent an ncRNA's sequence and secondary structure in the parameter space of a context-free grammar (CFG). Being a string defined on a special alphabet constructed from a CFG, it converts ncRNA alignment into sequence alignment with O(n^2) complexity. Moreover, all other existing techniques developed for strings or sequences such as multiple alignment, clustering, sequence database indexing can be applied to grammar strings with minor modifications. Our experiments have shown that grammar string based multiple sequence alignment competes favorably in consensus structure quality with several representative structural comparison tools. Availability: Source codes and experimental data are available at http://www.cse.msu.edu/~yannisun/grammar-string.

Soft Biometric Traits For Continuous User Authentication

2011-05-12

Authors: Koichiro Niinuma, Unsang Park, Anil K. Jain

Towards High Performance Security Policy Evaluation

2011-05-12

Authors: Qiang Wang, Fei Chen, Alex X. Liu, Zhiguang Qin

The Enterprise Privacy Authorization Language (EPAL) is a formal language for specifying fine-grained enterprise privacy policies. With the adoption of EPAL, especially in web applications, the performance of EPAL policy evaluation engines becomes a critical issue. In this paper, we propose Eengine, an engine for efficient EPAL policy evaluation. Eengine first converts all string values in an EPAL policy to numerical values. Second, it converts a numericalized EPAL policy specified as a list of rules following the first-match semantics to a tree structure for efficient processing of numericalized requests.

Online feature selection using group-LASSO

2011-05-12

Authors: Pavan K. Mallapragada, Rong Jin, Anil K. Jain

Given a pair of images represented using bag-of-visual-words and a label corresponding to whether the images are ``related''(must-link constraint) or ``unrelated'' (cannot-link constraint), we address the problem of selecting a subset of visual words that are salient in explaining the relation between the image pair. In particular, a subset of features is selected such that the distance computed using these features satisfies the given pairwise constraints. An efficient online feature selection algorithm is presented based on the dual-gradient descent approach. Side information in the form of pair-wise constraints is incorporated into the feature selection stage, providing the user with flexibility to use an unsupervised or semi-supervised algorithm at a later stage. Correlated subsets of visual words, usually resulting from hierarchical quantization process (called groups), are exploited to select a significantly smaller vocabulary. A group-LASSO regularizer is used to drive as many feature weights to zero as possible. We evaluate the quality of the pruned vocabulary by clustering the data using the resulting feature subset. Experiments on PASCAL VOC 2007 dataset using 5000 visual keywords, resulted in around 80% reduction in the number of keywords, with little or no loss in performance.

PDF

Evolution of Controllers for Mobile Ad Hoc Networks

2011-05-12

Authors: David B. Knoester, Heather J. Goldsby, Philip K. McKinley

This paper describes a study of the evolution of distributed behavior, specifically the control of agents in a mobile ad hoc network, using neuroevolution. In neuroevolution, a population of artificial neural networks (ANNs) are subject to mutation and natural selection. For this study, we compare three different neuroevolutionary systems: a direct encoding, an indirect encoding, and an indirect encoding that supports heterogeneity. Multiple variations of each of these systems were tested on a problem where agents were able to coordinate their collective behavior. Specifically, movement of agents in a simulated physics environment affected which agents were able to communicate with each other. The results of experiments indicate that this is a challenging problem domain for neuroevolution, and although direct and indirect encodings tended to perform similarly in our tests, the strategies employed by indirect encodings tended to favor stable, cohesive groups, while the direct encoding versions appeared more stochastic in nature.

Large Scale Hamming Distance Query Processing

2011-05-12

Authors: Alex X. Liu, Ke Shen, Eric Torng

Hamming distance has been widely used in many application domains, such as near-duplicate detection and pattern recognition. We study Hamming distance range query problems, where the goal is to find all strings in a database that are within a Hamming distance bound $k$ from a query string. If $k$ is fixed, we have a static Hamming distance range query problem. If $k$ is part of the input, we have a dynamic Hamming distance range query problem. For the static problem, the prior art uses lots of memory due to its aggressive replication of the database. For the dynamic range query problem, as far as we know, there is no space and time efficient solution for arbitrary databases. In this paper, we first propose a static Hamming distance range query algorithm called HEngine$^s$, which addresses the space issue in prior art by dynamically expanding the query on the fly. Second, we propose a dynamic Hamming distance range query algorithm called HEngine$^d$, which addresses the limitation in prior art using a divide-and-conquer strategy. Third, we present ways to convert $K$-nearest neighbor search problems and range query problems in other distance spaces to range query problems in Hamming space that can be solved by our HEngine algorithms. We implemented our algorithms and conducted side-by-side comparisons on large real-world and synthetic datasets. In our experiments, HEngine$^s$ uses 4.65 times less space and processes queries 16\% faster than the prior art, and HEngine$^d$ processes queries 46 times faster than linear scan while using only 1.7 times more space.

Periocular Biometrics in the Visible Spectrum

2011-05-12

Authors: Unsang Park, Raghavender Jillela, Arun Ross, Anil K. Jain

Periocular biometric refers to the facial region in the immediate vicinity of the eye. Acquisition of the periocular biometric does not require high user cooperation and close capture distance unlike other ocular biometrics (e.g., iris, retina, and sclera). We study the feasibility of using periocular images as a biometric trait. Global and local information are extracted from the periocular region using texture and point operators resulting in a feature set that can be used for matching. A number of aspects are studied in this work including (i) the effectiveness of the inclusion of eyebrows, (ii) the use of side information (left or right) in matching, (iii) manual vs. automatic periocular region segmentation, (iv) local vs. global feature extraction scheme, (v) fusion of face and periocular biometrics and (vi) usefulness of periocular biometrics with partially occluded face images. The experimental results show a rank-one recognition accuracy of 88.38% using 3,408 periocular images taken from 568 different subjects in the FRGC (version 2.0) database with the fusion of three different matchers.

Epigenetic Developers

2011-05-12

Authors: Juyang Weng, Matthew D. Luciw

Modeling brain-mind development, the epigenetic developer (ED) introduced here is a basic but general-purpose "building block" which seems to have the potential to develop simple or complex brains.

The Brain-Mind Puzzle in Each Reviewer’s Brain

2011-05-12

Authors: Juyang Weng

A General-Purpose Developmental Model of the Brain for Spatiotemporal Events in Complex Backgrounds

2011-05-12

Authors: Juyang Weng, Matthew D. Luciw

We present a general purpose model of the brain, called Self-Aware and Self-Effecting (SASE) model, after discussing the current lack of infrastructure for brain-scale research which also caused great challenges for any individual to adequately review the model here. Rooted in the biological genomic equivalence principle, our model proposes a general-purpose cell-centered in-place learning scheme to handle all levels of brain development and operation, from the cell level all the way to the brain level. It clarifies five necessary "chunks" of the brain "puzzle": development, architecture, area, space and time. The "development" chunk means that any practical brain, natural or artificial, needs to autonomously develop through interactions with the natural environments without any previously given set of tasks. The "architecture" chunk handles (1) complex backgrounds where the signal-to-noise ratio is at least smaller than 1 (< 0 db), or equivalently, more input components are irrelevant to immediate actions than those that are relevant; (2) abstraction, reasoning and generalization with any abstract and concrete contexts; (3) multiple sensory modalities and multiple motor modalities and their integration. The "area" chunk addresses the issue of feature development and area representation, without rigidly specifying what each neuron does. The "space" chunk deals with any practical foreground objects with any practical complex backgrounds, and includes conflicting invariance and specificity criteria for type, location, size, orientation, expression, etc. Learned context-dependent spatial attention is a key capability for dealing with all these conflicting spatial criteria. The "time" chunk indicates that the brain uses its intrinsic spatial mechanisms to deals with time, without dedicated temporal components. The model copes with practical temporal contexts, including the conflicting criteria of time warping, time duration, temporal attention, long temporal length, etc. Learned context-dependent temporal attention is a key capability for meeting all these conflicting criteria. The theory, formulations, detailed algorithms, and some related experimental results are given.

Adaptive Voice Stream Multicast over Low-power Wireless Networks

2011-05-12

Authors: Liqun Li, Guoliang Xing, Qi Han, Limin Sun

Low-power Wireless Networks (LWNs) have become increasingly available for mission-critical applications such as security surveillance and disaster response. In particular, emerging low-power wireless audio platforms provide an economical solution for ad hoc voice communication in emergency scenarios. In this paper, we develop a system called Adaptive Stream Multicast (ASM) for voice communication over multi-hop LWNs. ASM is composed of several novel components specially designed to deliver robust voice quality for multiple sinks in dynamic environments: 1) an empirical model to automatically evaluate the voice quality perceived at sinks based on current network condition; 2) a feedback-based Forward Error Correction scheme where the source can adapt its coding redundancy ratio dynamically in response to the voice quality variation at sinks; 3) a Tree-based Opportunistic Routing (TOR) protocol that fully exploits the broadcast opportunities on a tree based on novel forwarder selection and coordination rules; and 4) a distributed admission control algorithm that ensures the voice quality guarantees when admitting new voice streams. ASM has been implemented on a low-power hardware platform and extensively evaluated through experiments on a testbed of 18 nodes.

PDF

Fidelity-Aware Utilization Control for Cyber-Physical Surveillance Systems

2011-05-12

Authors: Jinzhu Chen, Rui Tan, Guoliang Xing, Xiaorui Wang, Xing Fu

Recent years have seen the growing deployments of Cyber-Physical Systems (CPSs) in many mission-critical applications such as security, civil infrastructure, and transportation. These applications often impose stringent requirements on system {\em sensing fidelity} and {\em timeliness}. However, existing approaches treat these two concerns in isolation and hence are not suitable for CPSs where system fidelity and timeliness are dependent of each other due to the tight integration of computational and physical resources. In this paper, we propose a holistic approach called {\em Fidelity-Aware Utilization Controller} (FAUC) for wireless cyber-physical surveillance (WCS) systems that combine low-end sensors with cameras for large-scale ad hoc surveillance in unplanned environments. By integrating data fusion with feedback control, FAUC can enforce a CPU utilization upper bound to ensure the system's real-time schedulability although CPU workloads vary significantly at runtime due to stochastic detection results. At the same time, FAUC optimizes system fidelity and adjusts the control objective of CPU utilization adaptively in the presence of variations of target/noise characteristics. We have implemented FAUC on a small-scale WCS testbed consisting of TelosB/Iris motes and cameras. Our extensive experiments on light and acoustic target detection show that FAUC can achieve robust fidelity and real-time guarantees in dynamic environments.

PDF

ZiFi: Wireless LAN Discovery via ZigBee Interference Signatures

2011-05-12

Authors: Ruogu Zhou, Yongping Xiong, Guoliang Xing, Limin Sun, Jian Ma

WiFi networks have enjoyed an unprecedent penetration rate in recent years. However, due to the limited coverage, existing WiFi infrastructure only provides intermittent connectivity for mobile users. Once leaving the current network coverage, WiFi clients must actively discover new WiFi access points (APs), which wastes the precious energy of mobile devices. Although several solutions have been proposed to address this issue, they either require significant modifications to existing network infrastructures or rely on context information that is not available in unknown environments. In this work, we develop a system called {\em ZiFi} that utilizes ZigBee radios to identify the existence of WiFi networks through unique interference signatures generated by WiFi beacons. We develop a new digital signal processing algorithm called Common Multiple Folding (CMF) that accurately amplifies periodic beacons in WiFi interference signals. ZiFi also adopts a constant false alarm rate (CFAR) detector that can minimize the false negative (FN) rate of WiFi beacon detection while satisfying the user-specified upper bound on false positive (FP) rate. We have implemented ZiFi on two platforms, a Linux netbook integrating a TelosB mote through the USB interface, and a Nokia N73 smartphone integrating a ZigBee card through the miniSD interface. Our experiments show that, under typical settings, ZiFi can detect WiFi APs with high accuracy ($<5\%$ total FP and FN rate), short delay ($\sim$ 780 {\em ms}), and little computation overhead.

PDF

Complexity Issues in Automated Model Revision Without Explicit Legitimate State

2011-05-12

Authors: Fuad Abujarad, Sandeep S. Kulkarni

Existing algorithms for the automated model revision incur an impediment that the designers have to identify the legitimate states of original model. Experience suggests that of the inputs required for model revision, identifying such legitimate states is the most difficult. In this paper, we consider the problem of automated model revision without explicit legitimate states. We show without the explicit legitimate states, in some instances, the complexity of model revision increases substantially (from P to NP-hard). In spite of this, we find that this formulation is relatively complete, i.e., if it was possible to perform model revision with explicit legitimate states then it is also possible to do so without the explicit identification of the legitimate states. Finally, we show if the problem of model revision can be solved with explicit legitimate states then the increased cost of solving it without explicit legitimate states is very small. In summary, the results in this paper identify instances of model revision where the explicit knowledge of legitimate state is beneficial and where it is not very crucial.

A Brain-Inspired Nework Reasons Abstractly with Time

2011-05-12

Authors: Juyang Weng

It is known that existing neural networks are largely one-short classifiers and they are not suited for continuously run context-based goal-directed reasoning that an Finite Automaton (FA) and all its probability variants (HMM, POMDP, Bayesian nets) can perform. Inspired by the brain architecture and the cerebral cortex, I show that a new type of recurrent network is not only duplicates all the functions of an FA, but also is more general. Thus, such a network performs context-based or state-based reasoning like a general purpose FA. The more general aspects of this network include: It is developmental in the sense that it has its internal representations which emerge from its "living" experience in the grounded physical world through interactions with its external environment and internal (inside the "skull") environment, without the need for a human to handcrafting its internal representation. Its architecture is sensor and motor driven (grounded), so that the network directly receives and outputs instances of numeric patterns, not just abstract symbols. Its building block is a dually optimal feature area called Lobe Component Analysis for the best use of network resource (size) and teaching resource (living time). Its space properties enable it to deal with goal-directed or context-directed attention to a small relevant part of the world in the presence of much irrelevant information (backgrounds). Its time properties extend the capability of dealing with infinitely long logic sequential operation by an FA to infinitely long spatiotemporal events in complex backgrounds. In particular, it deals with the challenging time warping issue like an FA and all its probability variants. The goal of such a network emerges from its motor end or be directed by the environment.

A Cross-Domain Privacy-Preserving Protocol for Cooperative Firewall Optimization

2011-05-12

Authors: Fei Chen, Bezawada Bruhadeshwar, Alex X. Liu

Firewalls have been widely deployed on the Internet for securing private networks. A firewall checks each incoming or outgoing packet to decide whether to accept or discard the packet based on its policy. Optimizing firewall policies is crucial for improving network performance. Prior work on firewall optimization focuses on either intra-firewall or inter-firewall optimization within one administrative domain where the privacy of firewall policies is not a concern. This paper explores inter-firewall optimization across administrative domains for the first time. The key technical challenge is that firewall policies cannot be shared across domains because a firewall policy contains confidential information and even potential security holes, which can be exploited by attackers. In this paper, we propose the first cross-domain privacy-preserving cooperative firewall policy optimization scheme. Specifically, for any two adjacent firewalls belonging to two different administrative domains, our scheme can identify in each firewall the rules that can be removed because of the other firewall. The optimization process involves cooperative computation between the two firewalls without any party disclosing its policy to the other. We implemented our scheme and conducted extensive experiments. The results on real firewall policies show that our scheme can remove as many as 49% of the rules in a firewall whereas the average is 19.4%. The communication cost is less than a few hundred KBs. Our scheme incurs no extra online packet processing overhead and the offline processing time is less than a few hundred seconds.

Towards High Performance Security Policy Evaluation

2011-05-12

Authors: Qiang Wang, Fei Chen, Alex X. Liu, Zhiguang qin

Quality-driven Volcanic Earthquake Detection using Wireless Sensor Networks

2011-05-12

Authors: Rui Tan, Guoliang Xing, Jinzhu Chen, Wenzhan Song, Renjie Huang

Volcano monitoring is of great interest to public safety and scientific explorations. However, traditional volcanic instrumentation such as broadband seismometers are expensive, power-hungry, bulky, and difficult to install. Wireless sensor networks (WSNs) offer the potential to monitor volcanos at unprecedented spatial and temporal scales. However, current volcanic WSN systems often yield poor monitoring quality due to the limited sensing capability of low-cost sensors and unpredictable dynamics of volcanic activities. Moreover, they are designed only for short-term monitoring due to the high energy consumption of centralized data collection. In this paper, we propose a novel quality-driven approach to achieving real-time, in-situ, and long-lived volcanic earthquake detection. By employing novel in-network collaborative signal processing algorithms, our approach can meet stringent requirements on sensing quality (low missing/false alarm rates and precise earthquake onset time) at low power consumption. We have implemented our algorithms in TinyOS and conducted extensive evaluation on a testbed of 24 TelosB motes as well as simulations based on real data traces collected during 5.5 months on an active volcano. We show that our approach yields near-zero false alarm and missing rates and less than one second of detection delay while achieving up to 6-fold energy reduction over the current data collection approach.

PDF

A Discriminative Model for Age Invariant Face Recognition

2011-05-12

Authors: Zhifeng Li

Aging variation poses one of the major problems to automatic face recognition systems. Most of the face recognition studies that have addressed the aging problem have focused on age estimation or aging simulation. However, research on age invariant face recognition is limited. Designing an appropriate feature representation and an effective matching framework for age invariant face recognition remains an open problem. In this paper, we propose a discriminative model to address face matching in the presence of age variation. In this framework, we first represent each face using two patch-based local feature representations, one based on scale invariant feature transform (SIFT) and the other based on multi-scale local binary patterns (MLBP). Since both SIFT-based features and MLBP-based features span a high-dimensional feature space, to reduce the feature dimensionality and avoid the over fitting problem, we use multi-feature discriminant analysis (MFDA) to process these two local feature spaces in a unified framework. The MFDA integrates two different random sampling techniques (random subspace and bagging) to improve the performance of linear discriminant analysis (LDA). By random sampling the training set as well as the feature space, multiple LDA-based classifiers are constructed and then combined to generate a robust decision via a fusion rule. Experimental results show that our approach outperforms a state-of-the-art commercial face recognition engine on MORPH, a large-scale public domain face aging dataset. We also compare the performance of the proposed discriminative model with a generative aging model. A fusion of discriminative and generative models further improves the face identification accuracy in the presence of aging. We have used 20,000 images from 10,000 subjects to evaluate the proposed system, which is the largest evaluation of facial aging study reported in the literature.

Towards Scalable Model Checking of Self-Stabilizing Programs

2011-05-12

Authors: Jingshu Chen, Fuad Abujarad, Sandeep Kulkarni

Existing approaches for verifying self-stabilization with a symbolic model checker have relied on the use of weak fairness. We point out that this approach has limited scalability. To overcome this limitation, first, we show that if self-stabilization is possible without fairness then cost of verifying self-stabilization is substantially lower. In fact, we observe from several case studies that cost of verification under weak fairness is more than 1000 times that of the cost without fairness. For the case where weak fairness is essential for self-stabilization, we identify two approaches for improving scalability: (1) decomposition and (2) utilizing the weaker version of self-stabilization, namely {\em weak stabilization}. In the first approach, designer partitions the program into components where each component satisfies its property without fairness.We show that the first approach enables us to verify Huang's mutual exclusion program for uniform rings with 31 processes (state space $10^{138}$) whereas without this approach, it was not possible to verify the same program with 5 processes (state space $10^{10}$). In the second approach, a weaker version of self-stabilization is verified. For Hoepman's ring-orientation program on odd-length ring, we show that it is possible to verify weak stabilization for 301 processes (state space $10^{181}$) whereas self-stabilization could not be verified for 9 processes (state space $10^{5}$) under weak fairness. Furthermore, one can utilize transformation algorithms to convert weak stabilizing programs to probabilistically stabilizing programs. Hence, for the case where it is not possible to verify deterministic self-stabilization, one can obtain the assurance provided by probabilistic self-stabilization at a significantly reduced cost. Finally, we also present 5 case studies to illustrate the scalability of stabilization with techniques suggested in this paper.

Separating overlapped fingerprints

2011-05-12

Authors: Fanglin Chen, Jianjiang Feng, Anil K. Jain, Jie Zhou, Jin Zhang

Fingerprint images generally contain either a single fingerprint (e.g., rolled images) or a set of non-overlapped fingerprints (e.g., slap fingerprints). However, there are situations where more than one fingerprints overlap on top of each other. Such situations are frequently encountered when latent (partial) fingerprints are lifted from crime scenes or residue fingerprints are left on fingerprint sensors. Overlapped fingerprints constitute a serious challenge to existing fingerprint recognition algorithms, since these algorithms are designed under the assumption that fingerprints have been properly segmented. In this paper, a novel algorithm is proposed to separate overlapped fingerprints into component or individual fingerprints. The basic idea is to first estimate the orientation field of the image with overlapped fingerprints and then separate it into component orientation fields using a relaxation labeling technique. We also propose an algorithm to utilize fingerprint singularity information to further improve the separation performance. Experimental results indicate that the algorithm leads to good separation of overlapped fingerprints that leads to a significant improvement in the matching accuracy.

Companion to “Automatically Generating Test Data from Conceptual Data Models”

2011-05-12

Authors: Matthew J. McGill, Laura K. Dillon, R.E.K. Stirewalt

Online Feture Selection

2011-05-12

Authors: Rong Jin

In this report, the problem of online feature selection is considered. In this problem instead of using all the features for classification, we require the classifier to use at most $B$ features for classification. We describe and analyze efficient algorithms for solving the mentioned problem. This problem can be considered as a special case in budgeted learning framework. Our work is related to budget online learning in that the number of support vectors is bounded by a predefined number. Also, our work is closely related to the online learning algorithm that aims to learn a classier that performs as well as the best subset of experts which is in contrast to most online learning work on prediction with expert advice that only compares to the best expert in the ensemble.