基于块自适应残差矢量量化的SAR图像压缩

SAR Image Compression Based on Block Adaptive Residual Vector Quantization

SAR图像的空间域压缩多采用矢量量化的方法进行。通常的块自适应矢量量化性能好但计算量较大,而残差矢量量化计算量小却在性能上有劣势。基于此,提出了一种新的基于块自适应残差矢量量化的SAR图像压缩算法。根据SAR幅值图像具有近似瑞利分布的特点对图像先进行基于瑞利分布的块自适应矢量量化,然后利用残差图像数据具有近似高斯分布的特点进行二次块自适应矢量量化。并且在残差矢量量化中进行了按照均方误差排序选择性压缩的改进,在压缩比不变的前提下有效提高了压缩系统性能。多幅图像的实验结果表明,相对传统的块自适应矢量量化和残差矢量量化,该算法在达到较高压缩性能和较低计算资源占用量以及较少存储空间占用上获得了有效的改进。

Vector quantization(VQ) is a commonly used technique in SAR image compression in spatial domain. Traditional block adaptive vector quantization(BAVQ) is usually good in performance but suffers from heavy calculations, while residual vector quantization(RVQ) has low calculation burden but is generally weak in performance. Combined with those characters, a new block adaptive residual vector quantization(BARVQ) method was proposed. Considering that SAR amplitude image distribution is approximately equal to Rayleigh distribution while residual image after first BAVQ encoding is quite near Gaussian distribution, BAVQ based on those distributions were used accordingly. Moreover, VQ encoding on selected blocks with higher mean squared error(MSE) of residual image was suggested to improve the BARVQ performance. Compared with BAVQ and RVQ, compression results on test images show that the new method achieves the improvement.