A Unified Multiresolution Framework for Automatic Target Recognition MIT AI Lab / LIDS A Unified Multiresolution Framework for Automatic Target Recognition Laboratory for Information and Decision Systems & Artificial Intelligence Laboratory Massachusetts Institute of Technology
Meeting Agenda Overview (John) MIT AI Lab / LIDS Meeting Agenda Overview (John) Multi-scale nonparametric analysis (Paul) Extensions to MNP approach (Alan) Discussion
Overview Convergence of two multiscale analysis methods Current Status MIT AI Lab / LIDS Overview Convergence of two multiscale analysis methods Multiresolution Auto-Regressive Models Irving, Willsky Multi-scale Nonparametric Progression de Bonet, Viola Current Status Continuing Efforts
Research Agenda Developing class probabilistic models MIT AI Lab / LIDS Research Agenda Developing class probabilistic models capturing variability Developing robust models and associated feature extraction algorithms saliency, minimal descriptions Developing computationally fast algorithms for large problems
Technical Approaches Sampling and nonparametric estimators MIT AI Lab / LIDS Technical Approaches Sampling and nonparametric estimators allows for modeling complex dependencies emphasis on multi-resolution approaches Sparse/robust representations phenomenology basis pursuit
Estimating dependencies across a multi-scale representation MIT AI Lab / LIDS Estimating dependencies across a multi-scale representation Structure is modeled as a process that evolves in scale Issues complexity of the dependency robustness with we can capture and exploit the dependency low conditional entropy within the representation
Multiresolution parent vector MIT AI Lab / LIDS Multiresolution parent vector Parent Vector V(x,y)={ } coarse fine
Extracting a Distribution of Parent Vectors MIT AI Lab / LIDS Extracting a Distribution of Parent Vectors
distribution condition MIT AI Lab / LIDS From a sample image we can infer a distribution of a multi-scale process... …from which synthesis sample registration discrimination distribution Likelihood Similarity example image segmentation denoising distribution condition super resolution
Current Status Emphasis has been on classification MIT AI Lab / LIDS Current Status Emphasis has been on classification testing almost exclusively with MSTAR public release data Completed 3-class testing on initial public release data BTR-70, BMP-2, T-72 2S-1, BRDM-2, D-7, T-62, ZIL-131, ZSU-23 models generated at 17 degrees, tested at 15 degrees over 2500 vehicle chips tested
MIT AI Lab / LIDS Models BMP2-C21 BTR70-C71 T72-132 Models for target vehicles were generated from example images: generated from vehicles with different numbers from the target vehicles only 10 examples, evenly distributed in heading angle measured at a depression angle of 17 degrees (targets were at 15 degrees)
Target vehicles BTR70-C71 Five target vehicles were used. MIT AI Lab / LIDS BTR70-C71 Target vehicles Five target vehicles were used. Vehicles which differed from the target class were included as confusion targets. There were 200-250 images in each class. BMP2-9563 BMP2-9566 T72-812 T72-S7
Confusion vehicles ZIL131 ZSU23 MIT AI Lab / LIDS ZIL131 ZSU23 Confusion vehicles Six additional confusion vehicles were used as well. 2S1 T62 BRDM2 D7
Multiscale Nonparametric Approach MIT AI Lab / LIDS BMP2-C21 BTR70-C71 T72-132 Multiscale Nonparametric Approach Template Matching
MIT AI Lab / LIDS Continuing Efforts Inclusion of T-72 variants from MSTAR public release IU Workshop Use of sub-aperture/sub-band features IU Workshop (sub-aperture) SPIE Aerosense Inclusion of attributed scattering center ?