Marlin L. Gendron

A Model of Clutter for Complex, Multivariate Geospatial Displays

Objective: A novel model of measuring clutter in complex geospatial displays was compared with human ratings of subjective clutter as a measure of convergent validity. The new model is called the color-clustering clutter (C3) model. Background: Clutter is a known problem in displays of complex data and has been shown to affect target search performance. Previous clutter models are discussed and compared with the C3 model. Method: Two experiments were performed. In Experiment 1, participants performed subjective clutter ratings on six classes of information visualizations. Empirical results were used to set two free parameters in the model. In Experiment 2, participants performed subjective clutter ratings on aeronautical charts. Both experiments compared and correlated empirical data to model predictions. Results: The first experiment resulted in a .76 correlation between ratings and C3. The second experiment resulted in a .86 correlation, significantly better than results from a model developed by Rosenholtz et al. Outliers to our correlation suggest further improvements to C3. Conclusions: We suggest that (a) the C3 model is a good predictor of subjective impressions of clutter in geospatial displays, (b) geospatial clutter is a function of color density and saliency (primary C3 components), and (c) pattern analysis techniques could further improve C3. Application: The C3 model could be used to improve the design of electronic geospatial displays by suggesting when a display will be too cluttered for its intended audience.

Vector map data compression with wavelets

Abstract : Wavelets and wavelet transforms can be used for vector-map data compression. The choice of wavelet, the level of decomposition, the method of thresholding, the height of the threshold, relative CPU times and file sizes, and reconstructed map appearance were investigated using the Wavelet Toolbox of MATLAB. Quantitative error measures were obtained. For two test vector-map data sets consisting of longitude and latitude points, compressions of 35 to 50 percent (1.5:1 to 2:1) were obtained with root-mean-square errors less than 0-003 to 0-01 deg longitude/latitude for wavelet packet decompositions using selected wavelets.

The Automated Change Detection and Classification Real-time (ACDC-RT) System

This paper presents an Automated Change Detection and Classification (ACDC) System, developed by the Naval Research Laboratory (NRL) and the Naval Oceanographic Office (NAVOCEANO), which aids analysts in performing change detection in real-time (RT) by co-registering new and historical imagery and using automated change detection algorithms that suggest imagery changes. In this paper, ACDC-RT components are described and results given from a recent change detection experiment.

WAVELET MULTI-SCALE EDGE DETECTION FOR EXTRACTION OF GEOGRAPHIC FEATURES TO IMPROVE VECTOR MAP DATABASES

Although numerous at smaller geographic scales, vector databases often do not exist at the more detailed, larger scales. A possible solution is the use of image processing techniques to detect edges in high-resolution satellite imagery. Features such as roads and airports are formed from the edges and matched up with similar features in existing low-resolution vector map databases, By replacing the old features with the new more accurate features, the resolution of the existing map database is improved. To accomplish this, a robust edge detection algorithm is needed that will perform well in noisy conditions. This paper studies and tests one such method, the Wavelet Multi-scale Edge Detector. The wavelet transform breaks down a signal into frequency bands at different levels. Noise present at lower scales smoothes out al higher levels. It is demonstrated that this property can be used to detect edges in noisy satellite imagery. Once edges are located, a new method will be proposed for storing these edges geographically so that features can be formed and paired with existing features in a vector map database.

Classifying sidescan sonar images using self organizing maps

Self organizing maps (SOMs) can be used for computer‐aided classification of objects found in two‐dimensional snippets of sidescan sonar images. SOMs are briefly discussed, including the choice of features or attributes as well as various types of input data. The inputs can be, for example, the data values themselves, either raw or processed images; the amplitudes of the Fourier transform coefficients of the data; the wavelet transform coefficients of the data; the energies of the horizontal, vertical, or diagonal wavelet coefficients; the autocorrelation of the data; the Hartley transform coefficients of the data; the cepstrum; the dimensions of the object; or the sonar bright spot and shadow character. Tabular results and two‐dimensional maps showing the groupings of measured and processed sidescan data are presented. Comparisons are made with human classifications of the same images. [Research supported in part by NRL‐ASEE Summer Faculty Research Program.]

Kalman filtering with neural networks for change detection in simulated sidescan sonar data

Sidescan sonars produce acoustical imagery which is used to detect bottom objects and characterize features of the seafloor. Change detection is a method that can be used to flag new bottom objects which were not detected during previous sidescan sweeps. Improvements in object detection and classification are critically needed to improve change detection methods. Adaptive filtering techniques may be used to identify objects known or previously marked from historical data, and flag new objects detected as changed. In this study Kalman Filter techniques will be used to estimate weights for a supervised feed‐forward perceptron neural net classifer. Both noise‐free and noisy data are considered. Preliminary results from these techniques using simulations that model sidescan sonar data sets will be presented.

Wavelets and neural networks for change detection in simulated sidescan sonar data

Recently recorded sidescan sonar data can be compared to historical data from the same area to detect any differences or changes. Such comparisons are complicated by the fact that the position is uncertain and the viewing angle or depth of the sidescan sonar may be different in the two data sets. Neural networks offer a possible technique to be used in the comparison procedure. Inputs to neural networks investigated here include wavelet transform coefficients of the data as well as the two‐dimensional sidescan sonar data itself. Several different neural network architectures and several different wavelet bases are investigated. Both noise‐free and noisy data are considered. Preliminary results for change detection using simulations that model simple historical and recent sidescan sonar data sets will be presented. [Research supported in part by NRL‐ASEE Summer Faculty Research Program.]

Wavelet denoising of sidescan sonar images

One important application of wavelet transforms is for noise removal or denoising. The effectiveness of this technique is influenced by the choice of wavelet used, the decomposition level, and the threshold (both amplitude and type). Thirty different wavelets, several allowable decomposition levels, and a range of appropriate thresholds are tested. Preliminary results will be presented of wavelet denoising applied to 2‐D acoustic backscatter imagery from a sidescan sonar in an attempt to improve the detection of bottom features. Comparisons with Fourier based filtering are also discussed. [Research supported in part by an NRL/ASEE Summer Faculty Fellowship and ONR.]

Change detection deconfliction process for sonar clutter items.

When resurveying a seafloor area of interest during change detection operations, an automated method to match found bottom objects with objects detected in a previous survey allows the surveyor to quickly sort new objects from old objects. Here we will demonstrate a software system that accomplishes change detection. The change detection system contains modules for automatic object detection by geospatial bitmap technique, object collocation, feature matching using shadow outlining, scene matching by control point matching, and visualization and filtering capabilities. Emphasis will be placed on the new elements of the system, namely, shadow outlining and optional spatial filtering.

Hierarchical clustering of historic sound speed profiles

Categorizing historical sound speed profiles stems from the desire to map spatial and temporal variability. Sound speed variability correlates with environmental phenomena and is an indicator of changes in sonar performance. Sound speed maps will assist in planning more efficient environmental survey operations such as conductivity, temperature, depth (CTD) collections. Intrinsic profile attributes estimated directly from the profile, such as the mean, variance, and derivative values, are used in the clustering process separately or in conjunction with extrinsic attributes such as location and ocean floor depth. Examples are used to demonstrate that the underlying spatial boundaries of the cluster groupings identify regions where sound speed profiles are consistent. The process is easily tailored to multiple clustering based upon the spatial and temporal scales of interest or on generic properties of the individual profile. The sensitivity of the cluster boundaries and group statistics to the addition of new profiles, or to changes in temporal and spatial scale, defines a new environmental characterization.