基于均匀网格编码量化的超光谱图像自适应压缩

ADAPTIVE COMPREESION OF HYPER-SPECTRAL IMAGES BASED ON UNIFORM TRELLIS-CODED QUANTIZATION

提出一种基于小波系数分类的超光谱图像压缩方法 .算法首先将各波段小波分解并将所得子带划分成子块 ,而后根据子块活动性将其分类 .在分类基础上 ,使用预测差分技术去除谱间冗余 ,此过程中分别求取各子类的预测系数以反映子带的局部相关性 .而后利用均匀网格编码量化方法来量化残差系数序列 ,最后使用自适应算术编码对量化码字进行熵编码 .为使编码器能在所有系数序列中最优地分配比特 ,本文提出一个基于序列统计特性和网格编码量化器率 -失真特性的比特分配算法 .实验证明该方法能高效地压缩超光谱图像 ,表现出优异的压缩性能 .

An approach for compression of hyper-spectral image based on classification of sub-bands was proposed firstly. The wavelet decomposition was carried out and the sub-bands were partitioned into sub-blocks. Then sub-blocks were classified based on their activity. Based on classification, the algorithm uses prediction to remove the spectral redundancy, in which the algorithm computes the predictor for each class to reflect local correlation in sub-band images. Then the uniform trellis-coded quantization is used to quantize the error images. At last, entropy encoding of the quantized codeword is performed by adaptive arithmetic encoding. To optimally allocate bits through all series of coefficients, an algorithm for bit allocation based on statistic characteristic of the series of coefficients and R-D characteristic of trellis-coded quantizer was proposed. The experiments show that the approach can efficiently compress hyper-spectral remote sensing images, and the excellent performance of the proposed algorithm is demonstrated.