A Context-Based, Signature-Predictive Classifier for Automated, Robust Material-Encoded Textures

JRM proposes a novel context-dependent, signature-based approach to improve the automation and quality of material-classification of imagery and arbitrary n-channel remote-sensed data. In the Phase I, JRM will design three key components: (1) an automated N-channel signature-predictive material classification algorithm, (2) a pixel-material spatial-correlation refinement algorithm and (3) an iterative context adjustment algorithm. The signature-predictive classifier will improve upon the current-art automated material DHR-fitting methods with phenomenology-based predictions of the at-aperture material spectral radiance that actually drive pixel intensity. This algorithm will use JRM's proven SigSim spectral signature prediction library, 0.2-25 um material property databases, improved fitting algorithms, and user-supplied sensor, environment and image-truth to produce likely pixel-material and mixture hypotheses and associated confidence metrics. The spatial-correlation refinement algorithm will reduce each pixel to a single, most-likely hypothesis with nearest-neighbor analysis methods. The iterative context adjuster will permit user re-finement under a different set of contexts. JRM will demonstrate feasibility of the Phase I designs in a proof-of-concept demonstration that produces a material-classified texture from an arbitrary photo-texture into a material-encoded texture, and simulating environmentally-dependent, radiometrically-correct OTW visual, NVG, FLIR and SAR outputs with it's VixSen real-time sensor simulation software from the resultant material-encoded texture.