Fast Vector Quantization Algorithm for Hyperspectral Image Compression

Summary form only given. HyperSpectral Image ( HSI ) has huge data volume, which brings a lot of difficulties to storage and transmission. VQ based HSI compression methods yield promising results. Generalized Lloyd Algorithm (GLA) is a classical codebook training method, but it takes a lot of computing resources. The dimensionality of HSI is more than one hundred. Those bands contain a lot of redundancy information and a lot of bands can be removed without loss of crucial information. It´s proved that the high dimensional space is mostly empty, which implies that multivariate data in R can be represented by a lower dimensional structure without losing significant information. So we can find a low dimension structure without losing significant information, which is good enough to represent the original high dimension. The main VQ computing task executes in the low dimensionality, so the VQ computing time can be dramatically reduced. DCT is one of feature transformation methods. We choose the most important several components, which is the good representation of the original data. Then the reduced date can be used in VQ algorithm. A new direction of fast algorithm for VQ based HSI compression has been proposed. It makes use of the fact that high dimensional space is mostly empty, so VQ based compression can be done at the lower dimensional subspace to reduce the computational complexity. The algorithm can be combined with other fast VQ algorithm to achieve less computational complexity. It can be easily applied to existing fast methods to further reduce computation time.