Information Fusion Technical Area Overview & Applications Joseph A Karakowski jkarakowski@caci.com (732)460-7752 November 16, 2011 CACI Private Data; Do Not Copy or Distribute What we will cover.. the “important” parts…..about fusion CACI Private Data; Do Not Copy or Distribute Agenda • Background Technology – Fusion Definition – Fusion Models • Fusion Technology Sector Applications – Military – Medical & Non-Military • Personal Fusion Areas (Optional) Questions to be Answered: What is Fusion Technology and it’s basis? What are some example fusion applications in specific markets? CACI Private Data; Do Not Copy or Distribute Some Fusion Definitions… …the process of combining data/information to estimate or predict the state of some aspect of the world (Bowman) …the process of utilising one or more data sources over time to assemble a representation of aspects of interest in an environment (Lambert) …series of processes performed to transform observational data into more detailed and refined information, knowledge, and understanding (USArmy) …everything is Connected… a “Global Graph” portrays the connected world; graph nodes are the entities; graph links are the actions or relationships (Walsh) CACI Private Data; Do Not Copy or Distribute What is Fusion? Technical methods/processes which supports, through cognitive/perceptual modeling, the solution of a class “Difficult Problems” These are a first scientific step to solve these classes of problems, which have not been solvable, up to this time. Implementation of these processes using information technology, has been moving forward for the last 25+ years, and will probably continue for many more years… Some Typical characteristic of Difficult Problems: • Multiple Goals • Complexity, with large numbers of items, interrelations and decisions • Dynamic, time considerations • Cognitive/perceptual problem solving CACI Private Data; Do Not Copy or Distribute “The Blind Man & the Elephant” Question: What is an Elephant? CACI Private Data; Do Not Copy or Distribute The “Fusion Elephant” It’s a Cognitive/ Perceptual Process! Question: What is Fusion? Its Biometric Apps! It’s Intelligence Apps ! It’s the JDL Model ! A State Prediction Problem! CACI Private Data; Do Not Copy or Distribute It’s a Global Graph ! A More Realistic “Fusion Elephant” Nuclear This is a new technology, and much RD&E remains CACI Private Data; Do Not Copy or Distribute Major Fusion Process Models Many Definitions…and many more models have been proposed and built! • Joint Directors of Laboratories Model (JDL)* [1986-Pres] • Transformation of Requirements for Information Process (TRIP) Model [2000-?] • Visual Situation Assessment Model () [1997] • Salerno SA Model [2001-Pres] • “Graph” Fusion Model [2005-Pres] • Contextual Fusion Model * [2009-Pres] • There are many others…. CACI Private Data; Do Not Copy or Distribute DFG Functional Model (JDL Model) Process Refinement Level 4 Human/ Computer Interaction Source Input Preprocessing/ Predetection Fusion Level 0 Single Object Refinement Level 1 Location; Attributes Behavior; Class; ID Situation Refinement Level 2 Aggregate object refinement Implications/ Threat Refinement Level 3 Situation Intent; interpretation Vulnerability Courses of Action Database Services Relatively static a priori Knowledge Dynamic Situation Database CACI Private Data; Do Not Copy or Distribute Richard Antony, in DFG Meeting Minutes, W. Doig, 14 March 1997 Antony & Karakowski Contextual Fusion Model (CFM) 2009 Conceptually organized along three(related) dimensions: (entity, context1, context2) AKA “Triple” “Fusion” ..an “assessment” operation between pairs of Triples: Lead to 8 fundamental classes of fusion operations “Fusion as a Process” exhausts all possible “assessment” combinations or fusion in a Triple; the result is a set of discovered concepts & relations from the fusion of the pairs of Triples. This provides a rich discovery space within an existing knowledge source CFM explicitly fuses diverse context with specific basic entities, all within a computational JDL model framework, resulting in a testable, expandable, and general fusion model CACI Private Data; Do Not Copy or Distribute Contextual Fusion Model Context & Content • Information for fusion requires both context and entity • “Entity” is the specific unit of information, or node in a graphical representation • Context allows perception of an Entity with respect to the information of interest – Context gives meaning to an Entity’s “information” – Context is required before an information entity has any meaning – Context must be an integral part of the fusion process & process model, its computation paradigm Context is knowledge that enhances the more complete understanding of a specific entity of interest and the desired resultant objective information product (s) 12 CACI Private Data; Do Not Copy or Distribute Two Entities within a graph, with Two sets of Two Entities, as their Contexts Entities as Context E6/C6 E5/C5 E4/C4 E3/C3 Graph Entity-Entiy Relations Entity Graph Nodes E1 E2 Green = Entity CACI Private Data; Do Not Copy or Distribute Eight Canonical Fusion Forms Intelligence Traditional Tracking/Correlation Application Fusion Operation = fusion of two entities with associated context Fusion Operation = (entity A, location A , time A) x (entity B, location B, time B) Contextual Dimensions Level 1 Fusion Form Individual Entity Location Time 1 Similar Similar Similar 2 Similar Similar Dissimilar 3 Similar Dissimilar Similar 4 Similar Dissimilar Dissimilar 5 Level 2 Fusion 6 7 8 Dissimilar Dissimilar Dissimilar Dissimilar Similar Similar Dissimilar Dissimilar Similar Dissimilar Similar Dissimilar Similarity/Dissim Assessment Op Example Concept of Fusion Result formed for Location and time context Fuse multiple, similar information sources; tracking using E-sensors Static similar object Infeasible condition – inconsistency; a truth maintenance signal Entity tracking/ tracking using both (slower) Esensors and messages Association / correlation of possible action as colocated Entities Association / correlation of entities based on same location only With prior Communication info: Potential Entity comm link (cell phone, chat, email) Multiple different entities at different times- further data mining and other FF conceptual analyses may be indicated CACI Private Data; Do Not Copy or Distribute 14 Simple Physical Context Example - Voltage If a voltage(entity) is viewed by itself without any context, we just “see” a value, either static, “semantic” or If voltage has added the context of apparently varying time, “signals” are created, with the Voltage field of electrical electronic engineering and associated signal analysis…. Note the huge information content difference between the entity of “voltage” and the addition of the context “time” and how context gives much more “meaning” to the entity (voltage) time . Human Entity, for example, a much more complex entity…. This is like a generalization from “humans” to “human signals” CACI Private Data; Do Not Copy or Distribute Eight Canonical Fusion Forms Source “Voltage” (just for fun) Fusion Operation = ( Source V1, location V1 , time V1) x (Source V2, location V2, time V2) Contextual Dimensions Level 1 Fusion Form V-Source Compare Location Time Concept of Fusion Result of Voltage Source for Location and time context 1 Similar Similar Similar Instantaneous single source value, at one place 2 Similar Similar Dissimilar Time varying single source value, at one place 3 Similar Dissimilar Similar 4 Similar Dissimilar Dissimilar 5 Dissimilar Similar Similar Instantaneous V-field for single source Instantaneous time-varying V-field for single source Instantaneous multiple source value, at one place Level 2 Fusion 6 Dissimilar Similar Dissimilar 7 Dissimilar Dissimilar Similar 8 Dissimilar Dissimilar Dissimilar Time varying / multiple sources value, at one place Instantaneous V-field for multiple sources Instantaneous time-varying V-field for multiple sources 16 CACI Private Data; Do Not Copy or Distribute Prior Art: Military Target Entity Model DRs of Organization to Organization DRs of Organizations to Events Level 2 Organizations DR: Discovered Relations thru contextual FFs DRs of Individuals to Organizations Level 2 DRs of Events to Events Events All Relations based on Location & Time Context Only DRs of Individuals to Individuals Individuals DRs of Individuals to Events Level 2 Other forms of discovery are possible; (I O, OE ) (EI ) eg CACI Private Data; Do Not Copy or Distribute 17 Fusion Applications CACI Private Data; Do Not Copy or Distribute Some Fusion Military Application Areas • • • • • • Intelligence Bio sensing/biometrics Can support at all levels: Hardware, Situation Awareness Software, Level 0 – Level 5 Imagery SIGINT(COMINT/ELINT) Tracking CACI Private Data; Do Not Copy or Distribute Non-Military Fusion Application Areas • • • • • • Networking/Cellular Homeland Security Medicine Chemistry Cognitive sciences …many others… CACI Private Data; Do Not Copy or Distribute Fusion “Topics” from a recent conference… CACI Private Data; Do Not Copy or Distribute Conference [Shortened-”C”] Index of Fusion Topics I • • • • • • • • • • • Camera Capability Acquisition Graph CBRN data fusion Cellular automata Centralized processing systems Challenge Problem Set Change detection Chemical plume Classification fusion Classification System Closest point approach • • • • • • • • • • • • Clustering algorithm Clutter Co-ranking Coalition formation Coalition operations Coarsening Coastal radar Cognitive Radio Networks Collaborative systems Collision mitigation Color Clustering …… CACI Private Data; Do Not Copy or Distribute Conference [Shortened-”C”] Index of Fusion Topics II • • • • • • • • • • Combination of belief • functions • Combinatory categorial grammar • Communication Decision • Communication failures • Complex object recognition • Compression • • Computer security • Conceptual graphs • • Conditional independence Confidence management Configuration Conflict analysis Confusion Conjunctive operator Connection Model Context Contradiction Convex optimization Convoy tracking Cooperative systems Coordinate registration CACI Private Data; Do Not Copy or Distribute Conference [Shortened-”C”] Index of Fusion Topics III • • • • • • • • • • • • • Coordination Correlation Course Of Action Covariance control Credal networks Credibility Crop modeling Cross correlation Cross-cueing Cubic Spline Curve Cued Sensors Cyber fusion Cyber-security From these three slides one can see both very specialized areas and much broader areas which currently utilize information Fusion technology CACI Private Data; Do Not Copy or Distribute Overview of Specific IF Apps from Selected market areas Military 1…Biometrics 2…Target Detection & Tracking 3…Chemical & Explosives 4…Image Fusion Medical Note: All these apps will fall somewhere in the fusion models and fusion definitions which I previously described. 5…Breast Cancer 6…Radiology Non-Military 7…Dept of Homeland Security 8…Cyber Security Summaries of specific fusion papers follows… CACI Private Data; Do Not Copy or Distribute 1. Biometrics A Multibiometric Face Recognition Fusion Framework with Template Protection[1] • “A fusion framework.. which demonstrates how …algorithms that produce hard decisions can be combined with unprotected algorithms that produce scores or soft decisions” Improving the recognition of fingerprint biometric system using enhanced image fusion[2] •“approach to increase the verification and identification of fingerprint recognition. This was achieved by using … linear fusion techniques” Multimodal Eye Recognition[3] • “results show that the proposed eye recognition method can achieve better performance…, and the accuracy of…kernel-based matching score fusion methods is higher than PCA and LDA” Military & Commercial CACI Private Data; Do Not Copy or Distribute 2. Target Detection &Tracking Level 0-2 fusion model for ATR using fuzzy logic[4] • “use of fusion at the lowest levels has been demonstrated. …provides a structure for fusion of multispectral data at all levels” Long-duration Fused Feature Learning Aided Tracking[5] • “Our experiments indicate that the Long-term Hypothesis Tree algorithm, which solves the tracklet-to-tracklet association problem, can be used to strongly disambiguate a multitude of situations and is a more computationally efficient algorithm than previously proposed joint solutions” Military CACI Private Data; Do Not Copy or Distribute 3. Chemical & Explosives Fusing chlorophyll fluorescence and plant canopy reflectance to detect TNT contamination in soils[6] • “physiological response of plants grown in TNT contaminated soils and … to detect uptake in plant leaves…use remote sensing of plant canopies to detect TNT soil contamination prior to visible signs” Sensor data fusion for spectroscopy-based detection of explosives[7] • “Multi-spot fusion is performed on a set of independent samples from the same region…. Furthermore, the results … are fused using linear discriminators. Improved detection performance with significantly reduced false alarm rates is reported using fusion techniques” Military Market CACI Private Data; Do Not Copy or Distribute 4. Image Fusion Towards Visual-Data Fusion[8] • “Fusion for both data and visual processes are derived as specific transforms from human linguistic requests. Visual “understanding” occurs by human-directed perception of summarized pattern representations within a familiar frame of reference” An orientation-based fusion algorithm for multisensor image fusion[9] • “Gabor wavelet transform … to fuse visible images and thermal images; orientation-based fusion is superior to the results of multiscale fusion algorithms…and can be applied to multiple (more than two) image fusion” Military & Commercial CACI Private Data; Do Not Copy or Distribute 5. Breast Cancer Investigation of PET/MRI Image Fusion Schemes for Enhanced Breast Cancer Diagnosis[10] • “results indicate that the radiologists were better able to perform a series of tasks when reading the fused PET/MRI data sets using color tables generated by our new genetic algorithm, as compared to commonly used …schemes” Time of Arrival Data Fusion Method for TwoDimensional Ultrawideband Breast Cancer Detection[11] •“A new microwave imaging method is given for breast tumor detection using an ultrawideband (UWB) imaging system. By combining the time of arrival (TOA) measurements from different sensors, the presence and location of small malignant lesions can be identified” Medical CACI Private Data; Do Not Copy or Distribute 6. Radiology KNOWLEDGE BASED FUZZY INFORMATION FUSION APPLIED TO CLASSIFICATION OF ABNORMAL BRAIN TISSUES FROM MRI[12] • “automatically classify abnormal tissues in human brain in a three dimension space from multispectral magnetic resonance images such as TI-weighted. T2- weighted and proton density feature images. It consists of four steps: data matching. information modeling, information fusion and fuzzy classification” New Applications of Planar Image Fusion in Clinical Nuclear Medicine and Radiology[13] • Fusion of multiple modalities has become an integral part of modern imaging methodology, especially in nuclear medicine where PET and SPECT scanning are frequently paired with computed tomography(CT). Additional fusing of orthopedic radiographs with photographic images of the extremities.. Medical – Add CACI Private Data; Do Not Copy or Distribute 7. DHS Military & Non-Military Information Fusion for CB Defense Applications[14] • “With appropriate algorithmic approaches and appropriately resolved tradeoffs, information fusion can offer… the potential of reaching performance that would be difficult, if not impossible, to attain otherwise. Thus, information fusion represents a significant opportunity for the CB defense and homeland security realm” Decision-level Information Fusion to Assess Threat Likelihood in Shipped Containers[15] •“details an approach to the decision-level fusion of disparate information to produce an assessment of the presence of a threat in a shipping container” Homeland Security Fusion Application of STEF[16] •“fusion system provided sufficient actionable intelligence that could have stopped a .. realistically staged terrorist attack on a US civilian target. …provided sufficient information to allow .. arresting the mastermind of the plot, as well as other key individuals and detaining the lower level individuals in his network, including the suicide bomber” CACI Private Data; Do Not Copy or Distribute 8. Cyber Security Application of the JDL Data Fusion Process Model for Cyber Security[17] • “explores the underlying processes identified in the Joint Directors of Laboratories (JDL) data fusion process model and further describes them in a cyber security context” Military & Non-Military CACI Private Data; Do Not Copy or Distribute We have covered “the more important parts”…a warning CACI Private Data; Do Not Copy or Distribute Summary & Closing Comments • Short background of Fusion Technology & Models/Contextual Fusion Model • Few examples of Fusion R&D / Apps • A lot was left out …. CACI Private Data; Do Not Copy or Distribute Backups CACI Private Data; Do Not Copy or Distribute Some of My Personal Fusion RDE Areas • UGS tracking/ID L0/L1 • RADAR ID – Signal processing L0/L1 / Fuzzy Expert – Confirmation/Disconfirmation • Voice Fingerprint ID biometrics L0/L1 • Visual fusion L0-L2[*] • Semantic/contextual unstructured information -understanding & discovery L1-L3[*] • Contextual Fusion System[2006-2010] • General Context Fusion Model [2011-?] CACI Private Data; Do Not Copy or Distribute Some Fusion Publications • • • • • • • • • • • Karakowski, J.A., “An Application of Text-Independent Speaker Recognition to High Speed Voice Surveillance”, Wide Area Surveillance Symposium, Office of Nat’l Drug Control Policy/Counter Drug Technology Assessment Center(1993) Karakowski, J.A., “Text Independent Speaker Recognition using A Fuzzy Hypercube Classifier”, ICASSP97(1997) Karakowski, J.A., “Towards Visual Fusion”, Invited Paper, Georgia Tech(1998). Antony, R. T. and Karakowski, J. A., “Service-Based Extensions to the JDL Fusion Model,” SPIE Defense Security and Sensing Conference (March 2008). Antony, R. T. and Karakowski, J. A., “Fusion of HUMINT & Conventional Multi-Source Data,” National Symposium on Sensor and Data Fusion, Session SC04 pp. 1-16 (07). Antony, R. T. & Karakowski, J. A., (2007) “Towards Greater Consciousness in Data Fusion Systems,” MSS National Symposium on Sensor and Data Fusion, (June 07). Antony, R. T. and Karakowski, J. A., “First-Principle Approach to Functionally Decomposing the JDL Fusion Model: Emphasis on Soft Target Data,” Fusion (July 08). Antony, R. T. & Karakowski, J. A., “Discovery Tools for Soft Target Applications,” National Symposium on Sensor and Data Fusion(2009) Antony, R. T. and Karakowski, J. A., “First-Principles Mapping of Fusion Applications into the JDL Model,” SPIE Defense Security and Sensing Conference (April 2009) Antony, R.T, & Karakowski, J.A., “Multiple Level-of-Abstraction Tracking and Alias Resolution”, National Symposium on Sensor and Data Fusion(2010) Antony, R.T., & Karakowski, J.A., “Toward more Robust Exploitation of the Asymmetric Threat: Binary Fusion Class Extensions”, (April 2011) SPIE. CACI Private Data; Do Not Copy or Distribute