Research On Transform-based SAR Image Compression Technique

The Synthetic Aperture Radar(SAR) is an active image system and the SAR image is obtained by reflection of electromagnetic wave. Because of the advantages of all-time, all-weather, strong penetrating power, SAR is widely used in military and civilian construction. With the continuous improvement of the SAR image spatial resolution, the amount of data increases rapidly. This status not only results in the contradiction between the needs of limited channel capacity and the transmission of large amounts of data, but also brings difficulties for long-term storage and further application of SAR images. Therefore, designing efficient SAR image compression algorithm becomes an important research area for developing advanced SAR system.Based on the spatial structure correlation of SAR image and the correlation between the multi-polarimetric SAR images, this thesis designs efficient compression algorithms by3D-wavelet transform, adaptive multiscale transform, sparse representation with dictionary learning and super-resolution reconstruction. The detailed work is conculded as follows.(1) The spatial correlation, speckle noise and traditional wavelet based compression performance of SAR images have been analysed. It laid a theoretical foundation for selecting proper transformation with SAR image compression.(2) Due to that there are a large number of similar characteristics in surface of the SAR image, we reorder the sub-blocks of SAR image according to the similarity measured by weighted Euclidean distance to form3D array. Then3D wavelet transform is used to remove the correlation between adjacent pixels and also can remove the correlation between the sub-blocks. At last,3D-SPIHT coding are employed for further improvement of the SAR image compression performance. The experimental results show the SAR image compression algorithm based on blocks reordering and3D wavelet transform has superior compression performance.(3) SAR images consist of textures and speckle noise, leading to that the high frequency coefficients of Tetrolet still have large amplitude, and thus affects the sparse representation performance of SAR images seriously. Tetrolet Packet adaptive multiscale transform is designed for SAR image. At first, the high frequency coefficients are reordered, and then the high frequency sub-bands are decomposed using multi-Tetrolet transform according to a entropy based cost function. So a optimal Tetrolet tree structure can be found, and the energy of coefficients are concentrated with less direction informations in order to get better performance of SAR image compression. Experimental results show that the Tetrolet Packet transform has better non-linear approximation performance, and thus can reduce the number of reserved coefficients. Furthermore, we explore the SAR image compression effect with Tetrolet Packet and Tetrolet algorithms.(4) Acording to the similar feature has similar structural characteristics of the SAR images, a SAR image compression algorithm based on sparse representation with dictionary learning is proposed. The SAR image set which covers different type of features is prepared for dictionary training, the dictionary then is used for sparse representation. By considering the relativity among the multi-polarimetric SAR images, a compression algorithm based on sparse representation with multilevel dictionary is proposed, followed by efficient quantify and entropy coding strategy. Experimental results show that the proposal algorithm introduces a large number of a priori information, and effectively improves the compression performance of the SAR image.(5) SAR image has very broad application requirements, and different applications require different quality of the SAR images. The high resolution SAR image is downsampled to low resolution image for reducing the code stream. Then, this low-resolution SAR image is compressed by sparse representation scheme. Finally, a derivative feature based super-resolution algorithm is used to restore the high frequency information removed during the down-sampling process. The proposal algorithm retains a large number of edge structure information of the SAR image, so it can accord with subjective quality evaluation standards. With the development of the SAR image application, more assessment criteria consistent with subjective quality will be found. So, according to different distortion measure, more super-resolution reconstruction based SAR image compression algorithms can be designed.In conclusion, this thesis has presented seveal compression algorithms for SAR image based on3D-wavelet transform, adaptive multiscale transform, sparse representation and super-resolution reconstruction theory, and the provided experiments have demonstrated their effectiveness.