基于SPIHT树结构和SVM回归算法的图像压缩研究  

IMAGE COMPRESSION METHOD BASED ON SPIHT COEFFICIENT TREES AND SUPPORT VECTOR MACHINES

作　　者：尹汪宏[1] 李朝峰[1] 张俊本[1] 王正友[2] 

机构地区：[1]江南大学信息工程学院,江苏无锡214122 [2]江西财经大学信息管理学院,江西南昌330013

出　　处：《计算机应用与软件》2008年第9期41-44,共4页Computer Applications and Software

基　　金：国家自然科学基金(60665001)

摘　　要：提出一种结合SPIHT系数树的支持向量回归图像压缩方法。首先通过离散小波变换,然后融合SPIHT树结构,分解出小波系数,以一个系数树上的系数构成一个向量,采取SVM回归实现对DWT系数相关性的学习,使用较少的支持向量表示原始系数,从而实现图像压缩。原始图像经过小波变换,分解成不同尺度的多个子频带,最低子频带系数集中了大部分能量,对图像重构起决定作用,直接采用DPCM编码,所有高频子带数据进行SVM回归压缩,最后所有数据采用算术编码。实验表明,本算法可有效提高图像的压缩效率,与JPEG2000算法相比较,在压缩率较高时,信噪比明显高于JPEG2000。A novel image compression algorithm based on SVM regression and SPIHT coefficient trees is presented. Firstly, image data are transformed by discrete wavelet transform. With the combination of SPIHT coefficient trees and SVM regression, relativity among DWT coefficients can be learned. Fewer support vectors are used to express original coefficients to realize image compression. Original image is decomposed into several sub-frequency bands of different scales via wavelet transform. The lowest sub-frequency band converges a majority of energy, which is decisive to image reconstruction, and DPCM coding is applied. All the higher sub-frequency bands are compressed by SVM regression. Eventually, all the data are coded by arithmetic coding. Experimental results show that the proposed method can effectively improve the image compression rate, and when the compression rate is high, the PSNR of this algorithm is higher than that of JPEG2000 algorithm.

关 键 词：图像压缩 SVM回归 DWT变换 熵编码 

分 类 号：TN919.81[电子电信—通信与信息系统] TS803.2[电子电信—信息与通信工程]