Research On Hyperspectral Image Lossless Compression Based On Lifting Wavelet Transform

Hyperspectral images are of interest for a number of applications, including environmental monitoring, geology, meteorology, medicine and military surveillance. As 3D images, hyperspectral image have another dimensional spectrum information again on the foundation of common and two-dimensional picture. So the amount of data of remotely sensed hyperspectral image is huge. As the development of hyperspectral imager, the hyperspectral images data increase largely, and bring huge pressure on data transmission and storage. Because the remote sensing information is precious and we expect high exactitude of resumed information, we give more concern on lossless or nearly lossy compression. In this work we focus on lossless hyperspectral images compression based on three-dimensional integer wavelet transform. The research works of the thesis are listed as following:The correlation of hyperspectral remote sensing images is tested at first.The test contains the computation of spatial correlation and spectral correlation.The results show: compared with common images, hyperspectral images have abundant texture and insignificant spatial correlation .But The hyperspectral remote sensing images have strong spectral statistical correlation and spectral structure correlation . In the compression algorithm design, the central focus is to get rid of the spectral correlation.The proposed algorithms here are based on the wavelet transform. Wavelet theory has been a topic of research in application math and engineering science. It is an important breakthrough of mathematics after Fourier transform. Compared with traditional image coder, the Wavelet image coder can compress image effectively and produce easily the embedded bitstream. The hyperspectral image contains rich information and has a high information entropy,so it is difficult to obtain large compressed lossless ratio.After the decomposition of hyperspectral images by wavelet transform,the image energy is mainly concentrated in the low-frequency part,while the energy of the horizontal,vertical and diagonal part is fewer,therefore the use of wavelet transform to compress the hyperspectral image will get better result. This paper researches the math theory base of wavelet transform and the character of the wavelet transform applied on image compressions. For lossless compression, the lifting scheme constructs the reversible integer wavelet transform. There are some advantages of this method such as: simple moves and addition operation, quick speed, occupy a little memory. It is specially suited for the situation which is real-time, high speed code or lossless compression and benefits the hardware realization in the future.The paper analyze the embedded wavelet coding. That is arraying the value of the wavelet coefficients from large to little, which typify their Contribution to the image,then transmitting the important value first. The encoding process can stop at any time when required bit rate or distortion is achieved. Correspondingly,the decoding process can stop whenever necessary and the restoring image at the bitstream truncation can be getted.The analysis focuses on SPIHT(Set Partioning in Hierarchical Trees) and SPECK(Set Partitioning Embedded BloCK coder). SPIHT organizes the wavelet coefficients together in the form of zero-tree, using the similarity of the same direction of the sub-bands. While SPECK organizes the wavelet coefficients together in the form of zero-block, using the correlation of the internal coefficients. Both of the two algorithms are applied to realize the compression of a single band of the test image. The results are analyzed and compared and it can be concluded that the performance of the two algorithms is almost the same and related to the correlation of the image.The compression effect will be better when the image has higher correlation.SPIHT is not suitable for the images,which have too much high-frequency information.As the hyperspectral image is three dimensional, has both spatial correlation and spectral correlation. Three dimensional integer wavelet transform can act on it, that is one-dimensional wavelet transform in spectral dimension and two-dimensional wavelet transform in spatial dimension.The transform decreases redundancy both in spatial dimension and spectral dimension. It makes image energy mainly concentrate in the low-frequency part.Repeat the process in low-frequency part,the energy becomes more concentrated, so the entropy is decreased.Samulation result proves that image after three-dimensional transform has a lower entropy than the two-dimensional transform image.Then 3DSPIHT or 3DSPECK are applied to realize the lossless compression(Both of the two algorithms are the expansion of two-dimensional algrithms).The results are tested and analyzed.According to the analysis of hyperspectral image, correlation in spatial dimension and spectral dimension is Asymmetric,So traditional three-dimensional wavelet transform,which is symmetric,is not optimal transform to the hyperspectral image compression.In response to this problem, An efficient wavelet packet transform is used to reduce redundancy in spectral dimension:Multi-level one-dimensional integer wavelet transform in spectral dimension then Multi-level two-dimensional integer wavelet transform in every band. Compared with traditional wavelet transform, transform acts on the high-frequency part in spectral dimension.This makes wavelet coefficients which have large absolute value gather at the upper-left corner of data cube. That is to say ,the energy is more concentrative. The simulation shows that 3DSPECK based on this transform achieves higher performance both on the quality of restoring image and compression ratio,than 3DSPIHT and 3DSPECK base on traditional 3D integer wavelet transform.