Researches On Multispectral Image Coding

Multispectral images are of interest for a large number of applications, including environmental monitoring, geology, meteorology, medicine and military surveillance. As 3D images, multispectral images result in large sized data sets. The storage and transmission of large volumes of multispectral data have become significant concerns. Therefore efficient compression is required for storage and transmission. The main work and achievements of the author on multispectral image coding in recent years are systematically introduced in this dissertation. In this work we focus on some technologies for image compression and 3D embedded wavelet coding. The achievements in this work are as follows:1. An improved 3D set partitioning embedded block (SPECK) algorithm based on an optimized quadtree coding is proposed. By analyzing context of subblock in zeroblock coding, new algorithm reduces the bit outputs of the significant blocks and coefficients. To avoid complex zeroblock sorting, a new ordering's strategy of List of Insignificant Sets (LIS) are employed in new algorithm. The simulation shows that the performance of the modified embedded wavelet image coder is increased and a higher computing efficiency is obtained.2. A 3D zeroblock algorithm based on Karhunen-Loeve transform and wavelet transform for multispectral image compression is proposed. First, a 2D wavelet transform is applied to reduce the spatial redundancies. Next a Karhunen-Loeve transform (KLT) is used to remove the correlation between adjacent spectral bands. Finally a modified 3D zeroblock coding method is used to code the transformed coefficients. Numerical experiments on some multispectral images show that the proposed KLT/WT-3DSPECK algorithm outperforms either the KLT/WT-3DSPIHT algorithm or the 3DWT-based AT-3DSPIHT and 3DSPECK algorithms. Besides having high performance, the KLT/WT-3DSPECK algorithm supports progressive transmission.3. An asymmetric transform 3D set partitioning embedded block (AT-3DSPECK) is proposed. An efficient wavelet packet transform is used to reduce redundancies in the spectral dimensions. According to the distribution of, energy of the transformed coefficients, the 3DSPECK's 3D set partitioning block algorithm and the 3D octave band partitioning scheme are efficiently combined in the proposed AT-3DSPECK algorithm. To accelerate the speed, a fast algorithm of the optimal zeroblock sorting is applied. The simulation shows that AT-3DSPECK either achieves higher performance than AT-3DSPIHT, or gains more than 2 times faster than AT-3DSPIHT encoding.4. A arbitrary-shape ROI zeroblock coding algorithm (AT-3DSPECK-ROI) based on fast lossless mask transform is proposed. The new algorithm employs "bit lifting" to generate the lossless ROI mask. By placing ROI-associated bites in the higher bitplanes, the ROI is coded with higher priority. Compared with AT-3DSPIHT-R0I, 3DSPIHT-ROI and JPEG2000-ROI method, AT-3DSPECK-ROI achieves the best compression performance.5. A novel object-oriented coding ( OB-3DSPECK ) algorithm based on shape-adaptive wavelet lifting transform and 3D zeroblock coding is proposed, shape-adaptive wavelet transform (SA-WT) based on lifting scheme is only applied on the pixels inside object region. The number of coefficients after SA-WT is identical to the number of pixels in the original arbitrarily shaped visual object. The important features of new SA-WT are that complex boundary extension and convolution in filter transform is not required, and processing speed is faster. By analyzing the location of invalid block in the mask image, OB-3DSPECK algorithm cancels symbol outputs of invalid block and coefficients outside the object, and only outputs two type symbols to arithmetic coding codec. Because reducing a type symbol, the new algorithm obviously enhances the performance of coding.