Affect of Image Compression on a Synthetic Aperture Radar Automatic TargetRecognition Prescreener and the Relation to SAR Image Statistics

摘要

High resolution SAR imagery compressed using three compression algorithms; VectorQuantization (VQ), JPEG and a Wavelet based MRES algorithm. The restored compressed imagery was processed through a morphology based prescreening Automatic Target Recognition algorithm and the results plotted on ROC curves. Based on these curves MRES and JPEG were found to perform significantly better than VQ. Several image statistics were computed for the restored compressed and uncompressed imagery. A prescreener performance figure of merit (FOM) was computed for each compression ratio for each algorithm. The correlation between each statistic and the FOM was computed. The results show the entropy is the statistic which correlates best with the FOM. Spearman's rank coefficient was used to perform a non parametric analysis on the image statistics. The results confirmed the first results showing that entropy is the best predictor of prescreener performance. This analysis also showed that JPEG and MRES results are fundamentally different from VQ results. This is not unexpected since these two are based on transforms whereas VQ is a purely empirical technique.