Performance evaluation of integer to integer wavelet transform for Synthetic Aperture Radar image compression

摘要

An efficient and novel imagery compression system for Synthetic Aperture Radar (SAR) which uses integer to integer wavelet transform and Modified Set Partitioning Embedded Block Coder (M-SPECK) has been presented in this paper. The presence of speckle noise, detailed texture, high dynamic range in SAR images, and even its vast data volume show the great differences of SAR imagery. Integer to integer wavelet transform is invertible in finite precision arithmetic, it maps integers to integers, and approximates linear wavelet transforms from which they are derived. Considering in terms of computational load, compression ratio and subjective visual quality metrics, several filter banks are compared together and some factors affecting the compression performance of the integer to integer wavelet transform are discussed in details. Then the optimal filter banks which are more appropriate for the SAR images compression are given. Information of high frequency has relatively larger proportion in SAR images compared with those of nature images. Measures to modify the quantizing thresholds in traditional SPECK are taken, which could be suitable to the contents of SAR imagery for the purpose of compression. Both the integer to integer wavelet transform and modified SPECK have the desirable feature of low computational complexity. Experimental results show its superiority over the traditional approaches in the condition of tradeoffs between compression efficiency and computational complexity.