Joint Target Detection and Lossy Compression of Tactical Sensor Data

More reliance is being placed in autonomous platforms to collect sensory data which are then transmitted to central sites for further processing and analysis. Efficient transmission of image data is essential and often requires lossy compression, which introduces error into the original data. The effects on associated processing tasks, such as target detection and recognition, can be detrimental if significant target features are corrupted. W propose to investigate the fundamental links between statistical nonparametric detection methods and investigate the use of adaptive wavelet transform tools for automated target or region of interest detection. In particular, we will investigate the links between vector quantization and tree-structured statistical detection tools and between hierarchical zerotree wavelet codecs and wavelet-based statistical detection algorithms. Based on the results of the initial study, we will develop a hybrid algorithm which provides high compression ratios while preserving accurate target detection capabilities.