Search

The detection of hazardous targets frequently requires a multispectral approach to image acquisition and analysis, which we have implemented in a software system called MATRE (multispectral automated target recognition and enhancement). MATRE provides capabilities of image enhancement, image database management, spectral signature extraction and visualization, statistical analysis of greyscale imagery, as well as 2D and 3D image processing operations. Our system is based upon a client-server architecture that is amenable to distributed implementation. In this paper, we discuss salient issues and requirements for multispectral recognition of hazardous targets, and show that our software fulfills or exceeds such requirements. MATRE's capabilities, as well as statistical and morphological analysis results, are exemplified with emphasis upon computational cost, ease of installation, and maintenance on various Unix platforms. Additionally, MATRE's image processing functions can be coded in vector-parallel form, for ease of implementation of SIMD-parallel processors. Our algorithms are expressed in terms of image algebra, a concise, rigorous notation that unifies linear and nonlinear mathematics in the image domain. An image algebra class library for the C + + language has been incorporated into the our system, which facilitates fast algorithm prototyping without the numerous drawbacks of descrete coding.© (1995) COPYRIGHT SPIE--The International Society for Optical Engineering. Downloading of the abstract is permitted for personal use only.

In Part I of this two-part series, we described the adaptive double-gated filter (DGF), a nonlinear CS detector that is sensitive to target shape and statistical properties. In particular, we derived the DGFs parameters from knowledge of target shape, size, and statistics, as well as background statistics. Several variants of the DGF were presented different for target recognition regimes (e.g., smooth targets against rough backgrounds, and vice versa). An additional variant of the DGF, known as the adaptive triple-gated-filter (TGF) was presented which aids in target registration. In this paper, we analyze the errors inherent in the DGF and TGF. Our analyses emphasize the performance of the DGF in limiting conditions, including high noise and poorly defined targets. Additionally, we discuss implementational optimization of the DGFs performance and time complexity. Examples of DGFs application to a database of field images presented, with discussion of results stated in terms of probability of hits and rate of false alarms.

Center-surround (CS) fields have long been identified in the human visual system as having properties of edge enhancement that facilitate the detection, location, and tracking of small objects. Unfortunately, the automatic digital implementation of such capabilities is not straightforward, since digital computers lack the neural functionality and connectivity inherent in the retina, optic chiasm, geniculate nuclei, and visual cortex. In this paper, we discuss efficient techniques for implementing CS detectors in terms of a class of nonlinear filters. In particular, we emphasize image enhancement schemes that cue the registration of CS detectors over the target, thus rendering our filters implementationally feasible in terms of hit rate and computational cost. Examples are given of the CS detectors that we call the adaptive Double- Gated and Triple-Gated filters (DGF and TGF). Such filters account for spatial nonstationary input. We further show how the DGF and TGF are useful for object detection in multispectral imagery. Analyses emphasize error and computational cost in practical sensing scenarios.