Joint ATR-compression for FLIR and SAR images

An algorithm for joint automatic target detection (ATD) and compression is investigated in this thesis. The objective is to develop a joint optimized compression system to achieve the highest compression ratios possible, while achieving high target detection rates at the same time. The motivation is that it is commonly required to compress sensor data obtained from airborne platforms prior to transmission due to the limited bandwidth channels available. Therefore, a compression-transmission-decompression strategy is necessary to facilitate real-time transmission. In addition, fast image analysis capability onboard an airborne platform is often desired for the intelligent bandwidth compression (IBC) or to avoid possible threats in a timely manner.; The simple and efficient baseline ATR algorithm is investigated so as to embed it into the compression system, for which we used a multistage scalar quantizer with subband decomposition, for a jointly-optimized compression system both for ATR and compression. Thus, we introduce a multiresolution target predictor based on the CFAR-like sliding window detector and edge detection in the frequency domain.; A joint compression problem is investigated by exploiting the synergism between the two principles, compression and detection/classification. We also suggest the more computationally efficient target predictor associated with the jointly-optimized encoder that identifies a target based on the probability that a particular pixel belongs to a target. These estimates are used to allocate bits spatially to optimize detection/recognition performance. In effect, this allows the more important target regions of the image (i.e., target chips) to be coded with a higher SNR than clutter regions.; The ATR-compression system is implemented and applied to different types of imagery such as FLIR, SAR, and food processing images. Its performance is evaluated and compared to a standard compression system such as the JPEG standard and subband coding without considering target information.; Our result shows that the ATR-compression system can be used effectively for the compression of sensor data.