基于ε-SVR的嵌入式小波图像编码算法  

ε-SVR Based Embedded Wavelet Image Coding Algorithm

作　　者：唐国维[1] 张岩[1] 王苫社[2] 刘彦彤[1] 赵璨[1] 

机构地区：[1]东北石油大学计算机与信息技术学院,黑龙江大庆163318 [2]哈尔滨工业大学计算机科学与技术学院,哈尔滨150001

出　　处：《吉林大学学报（信息科学版）》2017年第1期76-84,共9页Journal of Jilin University（Information Science Edition）

基　　金：国家自然科学基金资助项目(61502094);黑龙江省自然科学基金资助项目(F20160002)

摘　　要：针对基于支持向量机的小波图像编码算法难以实现嵌入式特性问题,在小波域构建一种回归树结构作为回归基本数据集合,同时利用子带内和子带间小波系数的相关性,提出一种线性动态阈值选取方法,以利于逐次逼近并保证回归数据的均衡性,并基于选定的阈值动态选取ε误差参数对小波系数进行多次回归,保证了重要系数被优先编码,使压缩算法具有嵌入式特性,对获得的支持向量及其权重进行自适应算术编码。实验结果表明,在压缩比相近的情况下,重构图像的PSNR(Peak Signal to Noise Ratio)比同类算法提高1~3 d B。The problem that support vector machine based wavelet image coding is difficult to achieve embedded characteristics was studied. Firstly,a regression tree structure was constructed in wavelet domain to act as the basic regression data set,which can use the inner-subband and inter-subband correlation simultaneously.Secondly,a linear dynamic threshold selecting method was put forward to facilitate the successive approximation and to make the regression data to be harmonious. Thirdly,based on the threshold selected,the SVM（Support Vector Machine） error parameters ε was dynamically determined in order to achieve multiple regression to the wavelet coefficients; the significant coefficients could be encoded prior and the impression algorithm was endowed with embedded characteristics. Finally,the adaptive arithmetic coding method was used to encode the support vectors and their weights. The experimental results show that,compared with the current similar algorithms,the PSNR（Peak Signal to Noise Ratio） of the reconstructed image is improved by 1 - 3 d B.

关 键 词：图像压缩 小波变换 嵌入式图像编码 支持向量机 ε-支持向量回归机 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]