Neil C. Rowe

Natural-language retrieval of images based on descriptive captions

We describe a prototype intelligent information retrieval system that uses natural-language understanding to efficiently locate captioned data. Multimedia data generally require captions to explain their features and significance. Such descriptive captions often rely on long nominal compounds (strings of consecutive nouns) which create problems of disambiguating word sence. In our system, captions and user queries are parsed and interpreted to produce a logical form using a detailed theory of the meaning of nominal compounds. A fine-grain match can then compare the logical form of the query to the logical forms for each caption. To improve system efficiency, we first perform a coarse-grain match with index files, using nouns and verbs extracted from the query. Our experiments with randomly selected queries and captions from an existing image library show an increase of 30% in precision and 50% in recall over the keyphrase approach currently used. Our processing times have a median of seven seconds as compared to eight minutes for the existing system, and our system is much easier to use.

Optimal grid-free path planning across arbitrarily contoured terrain with anisotropic friction and gravity effects

The authors address anisotropic friction and gravity effects as well as ranges of impermissible-traversal headings due to overturn danger or power limitations. The method does not require imposition of a uniform grid, nor does it average effects in different directions, but reasons about a polyhedral approximation of terrain. It reduces the problem to a finite but provably optimal set of possibilities and then uses A* search to find the cost-optimal path. However, the possibilities are not physical locations but path subspaces. The method also exploits the insight that there are only four ways to optimally traverse an anisotropic homogeneous region: (1) straight across without braking, which is the standard isotropic-weighted-region traversal; (2) straight across without braking but as close as possible to a desired impermissible heading; (3) making impermissibility-avoiding switchbacks on the path across a region; and (4) straight across with braking. The authors prove specific optimality criteria for transitions on the boundaries of regions for each combination of traversal types. >

Natural-language processing support for developing policy-governed software systems

Organizations are policy-driven entities. Policy bases can be very large and the relationships between policies can be complex. In addition, policy can change on a frequent basis. Checking for gaps in policy or analyzing the ramifications of changing policy is necessary to both identify and rectify gaps or unintended policy prior to the policy base being refined into requirements for a system. A policy workbench is an integrated set of computer based tools for developing, reasoning about, and maintaining policy. A workbench takes as input a computationally equivalent form of policy statements. We have developed a prototype of a tool that maps natural language policy statements to an equivalent computational form. The authors describe the architecture of a natural language input-processing tool (NLIPT). It has an extractor, which generates a meaning list representative of the natural language input; an index-term generator, which identifies the key terms used to index relevant policy schema in the policy base; a structural modeler, which structures a schema for input; and a logic modeler, which maps the schema to an equivalent logical form. We experimented with a prototype of the extractor which successfully parsed a sample of ninety-nine Naval Postgraduate School security policy statements with ninety-six percent accuracy.

An efficient Snell's law method for optimal-path planning across multipole two-dimensional, irregular, homogeneous-cost regions

We are exploring a new approach to high-level optimal-path planning when homogeneous irregularly shaped regions of a plane have different traversal costs per unit distance. It is based on the simple idea that optimal paths must be straight in homogeneous regions, and so those regions need not be subdivided for path planning. Our approach uses optics anal ogies, ray tracing, and Snells Law and reduces the problem to an efficient graph search with a variety of pruning criteria. The time and space of our algorithm is O(V 2) in an intui tively average case, where V is the number of region vertices. In experiments in which space is held constant, an imple mentation of our algorithm was not only dramatically faster but also gave better cost solution paths than a representative implementation of the chief competing technique, grid-based wavefront propagation. This appears to be because of the quite different and considerably smaller search space required for our more intelligent algorithm.

Change detection for linear features in aerial photographs using edge-finding

The authors describe a system fur automatic change detection of linear features such as roads and buildings in aerial photographs. Rather than compare pixels, they match major line segments and note those without counterparts. Experiments show their methods to be promising for images of around 2-m resolution.

Defending Cyberspace with Fake Honeypots

Honeypots are computer systems designed for no purpose other than recording attacks on them. Cyberattackers should avoid them since honeypots jeopardize the secrecy of attack methods and it is hard to launch attacks from them. This suggests that a computer system might pretend to be a honeypot to scare away attackers, reducing the number of attacks and their severity. This could be done on ordinary computer systems as a kind of “vaccination” of those systems, to create what we call “fake honeypots”. After some background, we examine this idea from three perspectives. We develop a mathematical model of what would make an attacker go away. We report experiments with deliberate distortions on text to see at what point people could detect deception, and discover they can respond to subtle clues. We then report experiments with real attackers against a honeypot. Results show that attacks on it decreased over time (which may indicate that attackers are being scared away), irregular outages of the honeypot stimulated attacks, and other changes occurred in response to our manipulations. We conclude with some speculation about the escalation of honeypot-antihoneypot techniques.

Finding Optimal-Path Maps for Path Planning across Weighted Regions

Optimal-path maps tell robots or people the best way to reach a goal point from anywhere in a known terrain area, eliminating most of the need to plan during travel. The authors address the construction of optimal-path maps for two-dimensional polygonal weighted-region terrain, terrain partitioned into polygonal areas such that the cost per unit of distance traveled is homogeneous and isotropic within each area. This is useful for overland route planning across varied ground surfaces and vegetation. The authors propose a new algorithm that recursively partitions terrain into regions of similar optimal-path behavior, and defines corresponding “path subspaces” for these regions. This process constructs a piecewise-smooth function of terrain position whose gradient direction is everywhere the optimal-path direction, permitting quick path finding. The algorithm used is more complicated than the current path-caching and wavefront-propagation algorithms, but it gives more accurate maps requiring less space to represent. Experiments with an implementation confirm the practicality of the authors’ algorithm.

Solving global two-dimensional routing problems using snell's law and a search

Long-range route planning is an important component in the intelligent control system of an autonomous agent. Most attempts to solve it with map data rely on applying simple search strategies to high-resolution, node-and-link representations of the map. These techniques have several disadvantages including large time and space requirements. We present an alternative which utilizes a more intelligent representation of the problem environment. Topographical features are represented as homogeneous-cost regions, greatly reducing storage requirements. Then, the A* search strategy is applied to a dynamically created graph, constructed according to Snells law. Testing has shown significant speed improvements over competing techniques.

Marie-4: a high-recall, self-improving Web crawler that finds images using captions

Marie-4, an intelligent software agent-based Web crawler and caption filter, searches the Web to find image captions and the associated image objects. It uses a broad set of criteria to yield higher recall than competing systems, which generally focus on high precision.

Automatic caption localization for photographs on World Wide Web pages

Abstract A variety of software tools index text of the World Wide Web, but little attention has been paid to the many photographs. We explore the indirect method of locating for indexing the likely explicit and implicit captions of photographs. We use multimodal clues including the specific words used, the syntax, the surrounding layout of the Web page, and the general appearance of the associated image. Our MARIE-3 system thus avoids full image processing and full natural-language processing, but shows a surprising degree of success. Experiments with a semi-random set of Web pages showed 41% recall with 41% precision for the task of distinguishing captions from other text, and 70% recall with 30% precision. This is much better than chance since actual captions were only 1.4% of the text on pages with photographs.