基于单邻点多波段预测的高光谱图像无损压缩算法

提出了一种基于聚类-单邻点、多波段预测-熵编码的高光谱数据无损压缩方法。根据谱向特征,进行高光谱图像矢量聚类。对各个分类,采用单个空间位置邻点、多个波段作为预测数据,训练预测系数,进行三维预测。残差采用Golomb-Rice编码。实验证实了算法的有效性。

Lossless Compression of Hyperspectral Images Based on Single Neighbor Multi-Bands Prediction

Applications for hyperspectral image data are still in their infancy as handling the significant size of the data presents a challenge for the user community. Data compression becomes a key problem. Based on clustering, predicting with single neighbor and self position in multi-bands, and entropy coding, a lossless compression method of hyperspectral images is presented. According to spectral structure, the spectra of a hyperspectral image are clustered by pixels. In every cluster, single spatial neighbor and the same spatial position of the current pixels are used for prediction. Using neighbors in various directions, four predictors are achieved. For each spatial position, one of the predictors is selected to perform the three dimension prediction. The residuals are entropy-coded using the Rice coding. The achieved compression ratios are compared with those of existing methods. The results show that the algorithm is an efficient method.