Research on Hyperspectral Image Compression Method Based on Information of Interest
|School||Harbin Institute of Technology|
|Course||Information and Communication Engineering|
|Keywords||Hyperspectral image information of interest hierarchical compression method image evaluation|
With much richer information and higher spectral resolution than multispectral image, hyperspectral image could resolve many problems that multispectral image could not do. But higher spectral resolution is accompanied by a huge volume of image data, which will result in excessive computing time and data complexity for transmission and storage, so it is necessary to compress hyperspectral image. In the most classical compression methods, low frequency information are reserved firstly, high frequency information are usually discarded which is insensitive to us. Because of limit of spatial resolution, some very important information for the application of hyperspectral image that we are interested in is usually belong to high frequency. So a new compression method for hyperspectral image is necessary to carry out which can preserve the information that we are interested in.In this thesis, we do some researches on hyperspectral image compression based on the protection of important information. Four main researches are done in this thesis.1) Analyzing hyperspectral image spatial and spectral correlation and features of our interested information. The strongly correlation is the base of hyperspectral image compression. Features of our interested information include Moments features and spectral absorption index (SAI). They are theoretical foundation for us to extract our interested information.2) Extraction and marking of our interested information. The main extraction methods are spectral derivative and spatial morphological filtering. Marking method is minimum enclosing rectangle (MER). Experiment results show that our method could successfully extract our interested information and mark them.3) Hierarchical compression based on the protection of our interested information. In this part, we classify hyperspectral image into different levels, according to their importance. To different level, we use different compression method. Our experiment based on AVIRIS hyperspectral image San Diego shows: our proposed classification prediction method could construct our most important information with few coding bits. And our bit-plan shifting method could better protect our interested information than usual classical compression methods at the same conditions. 3D-SPIHT method is used to the common information in order to ensure our compression efficiency.4) Hyperspectral reconstruction image evaluation based on target detection. We reconstruct hyperspectral image at 0.1 and 0.2bpp (bits per pixel), using our proposed compression method. The experiment result shows that our reconstruction image could gain better performance in target detection.In this thesis, in order to improve hyperspectral image target detection performance, we proposed a hierarchical compression method based on the protection of information. The experiment result shows that in the same condition, our proposed compression method could improve reconstruction image application performance.