Data and Applications Security Developments and Directions Dr. Bhavani Thuraisingham The University of Texas at Dallas Lecture #20 Secure Multimedia Data Management and Geospatial data management March 19, 2008

Outline l Multimedia Data Management Systems l Security l Secure Geospatial data management

Why Multimedia Data Management System? l Need persistent storage for managing large quantities of multimedia data l A Multimedia data manager manages multimedia data such as text, images, audio, animation, video l Extended by a Browser to produce a Hypermedia data management system l Heterogeneity with respect to data types l Numerous Applications - Entertainment, Defense and Intelligence, Telecommunications, Finance, Medical

Architectures: Loose Integration Multimedia File Manager Metadata Module for Integrating Data Manager with File Manager User Interface Data Manager for Metadata Multimedia Files

Architectures: Tight Integration User Interface MM-DBMS: Integrated data manager and file manager MM-DBMS: Integrated data manager and file manager Multimedia Database

Data Model: Scenario Example: Object representation Object A 2000 Frames 4/95 8/95 5/95 10/95 Object B 3000 Frames

Multimedia Data Access: Some approaches l Text data - Selection with index features - Methods: Full text scanning, Inverted files, Document clustering l Audio/Speech data - Pattern matching algorithms l Matching index features given for searching and ones available in the database l Image data - Identifying geometric boundaries, Identifying spatial relationships, Image clustering l Video data - Retrieval with metadata, Pattern matching with images

Metadata for Multimedia l Metadata may be annotations and stored in relations - I.e., Metadata from text, images, audio and video are extracted as stored as text - Text metadata may be converted to relations by tagging and extracting concepts l Metadata may be images of video data - E.g., certain frames may be captured as metadata l Multimedia data understanding - Extracting metadata from the multimedia data

Storage Methods l Single disk storage - Objects belonging to different media types in same disk l Multiple disk storage - Objects distributed across disks l Example: individual media types stored in different disks l I.e., audio in one disk and video in another l Need to synchronize for presentation (real-time techniques) l Multiple disks with striping - Distribute placement of media objects in different disks l Called disk striping

Security Issues l Access Control l Multilevel Security l Architecture l Secure Geospatial Information Systems

Access Control for Multimedia Databases l Access Control for Text, Images, Audio and Video l Granularity of Protection - Text l John has access to Chapters 1 and 2 but not to 3 and 4 - Images l John has access to portions of the image l Access control for pixels? - Video and Audio l John has access to Frames 1000 to 2000 l Jane has access only to scenes in US - Security constraints l Association based constraints E.g., collections of images are classified

MLS Security

Example Security Architecture: Integrity Lock

Inference Control

Securing Geospatial Data l Geospatial images could be Digital Raster Images that store images as pixels or Digital Vector Images that store images as points, lines and polygons l GSAM: Geospatial Authorization Model specifies subjects, credentials, objects (e.g, points, lines, pixels etc.) and the access that subjects have to objects l Reference: Authorization Model for Geospatial Data; Atluri and Chun, IEEE Transactions on Dependable and Secure Computing, Volume 1, #4, October – December 2004.

Framework for Geospatial Data Security (Joint with UCDavis and Purdue U.)

Example of several GIS repositories and GIS themes/layers for Northern California (Gertz, Bertino, Thuraisingham) Assume a single GIS data repository that manages information about parcels (being the basic units of geography for local government) and cadastre, including land use and zoning, environmental areas, and municipal utility services. Such type of repository is typically used by public sector staff to assist property owners and to support emergency, fire, and police operations. The latter type of usage includes identifying property structures and owners. Parcel maps in particular can be useful to do damage assessment after a disaster.

Example (Continued) They are also an important access point during emergencies for linking data from different GIS repositories. While such types of geospatial are used to serve the public, e.g., through Web-based interfaces, not all data layers are made publicly available. For example, property owner information is not publicly accessible A similar separation of public and private GIS data can be made for other types of themes. For example, environmental theme layers do not make information about locations of endangered species or nesting sites public. Based on this type of separation of GIS data, the following question arises: “What security mechanisms are used to specify and enforce different types of access to data in a single GIS repository?” In particular, “What provisions do GSI data managers have to (1) give public sector staff only access to GIS data relevant to their function, and (2) ensure that no sensitive geospatial data (e.g., parcel owner information) is made publicly available?” Ideally, GIS repositories should provide access control models and techniques similar to those developed for traditional (relational) databases. However, the diversity of geospatial data (feature- based versus field-based) and the complexity of feature-based geospatial data complicate a coherent and uniform access control model.

Policy Example (Bertino, Gertz, Thuraisingham) Deny/allow policies with flexible granularity, grouping mechanisms for protected objects, and space-related access restrictions. Deny/allow policies will be supported through the use of positive/negative authorizations; negative authorizations are crucial in order to support exceptions, by which, for example, an authorization is assigned to all objects in a set but one. In our context this paradigm is complicated by the larger options that we provide for denoting protected objects and by the presence of different object representations and dimensions. The main mechanism that we provide to support flexible grouping is based on the notions of object-locator and spatial window. An object-locator is a query expression that may include predicates against properties of feature types, metadata and provenance data. Predicates may also refer to topological relationships holding among the data objects, such as Within and Touches. An example of a policy using Touches is the one allowing a subject, which has access to information on a particular land parcel, to access information about all adjacent land parcels. The query expression may also include a projection component to specify an object representation and components. A spatial window is simply a spatial region in the reference space and denotes the set of object that are inside the boundary of the region. By combining such two mechanisms, one can specify sets of objects such as “all shelters occupying an area greater than 3000sf in Montgomery County”; in such case Montgomery County represents the authorization window. The use of spatial windows is particularly important to

Policy Example (Continued) Active policies. These are policies that when applied to a protected object perform certain transformations on the object, before returning it to the requester. Two relevant classes are the filtering policies and the obfuscating policies. Filtering policies refer to policies that filter out some portions of the objects before returning them to the users. These policies are directly supported by our object locator mechanisms. Obfuscating policies These policies act like filter policies except that they do not simply select objects but perform possibly complex computations on the feature(s) to be returned. Typical examples include computing a lower resolution image, and distorting some vector data (but preserving topological relationships). One can even specify policies that return incorrect data (e.g., as a honey pot in the context of misuse detection). In our model these policies are supported by the projection component, suitably extended with the possibility of invoking functions, of the object locator. We will provide a library including a variety of functions to support obfuscating policies.

Policy Example (Concluded) Context-dependent access control policies. Under such policies, information from the environment is taken into account by the access control module when taking decisions about access requests. Typical contextual information includes time and subject location. Subject location information is used to specify policies allowing a subject to access a resource only if the current location of the subject verifies certain spatial constraints. Context-dependent access policies will be supported by the introduction of a context component, as part of authorization rules, and by attribute-based specification of subjects in authorization rules. Event-based access control policies. Event-based access control policies are novel and are based on the idea that policies can be enabled/disabled depending on the occurrence of specified events. Events can include data modifications, very much like in database triggers, or application-dependent events, such as an emergency. We notice that current sensor networks and intelligent appliances make it very easy for a computer system to detect events arising in the environments. Our model will take advantage of such capabilities.

Policy Language l Take existing geospatial language/model and extend for security - E.g., GML l Take a security model/language and extend for geospatial - E.g, XACML has been extended to Geo-XACML l Develop from scratch - GRDF, Secure GRDF (developed at UTDallas by Alam Ashraful for PhD research)

The strength of RDF lies in the ease of composition with which RDF based formalisms can be integrated with other similar languages. On the Semantic Web, the goal is to minimize human intervention and to make way for machines to perform rule based automated reasoning. We are developing GRDF for geospatial data representation Why not use GML? - same reasons for using RDF and not XML – semantics Secure GRDF – security extensions for GRDF Geospatial Semantic Web: GRDF

Directions l Little research on Multimedia and Geospatial data security l Digital watermarking is getting attention l Our focus at UTD is to develop a secure geospatial semantic web l We have developed a system called DAGIS and demonstrating secure interoperability