摘要
给出了一个在低码率图像描述中应用最优支持向量(SV)描述的编码算法,提出一个训练样本集缩减策略,在支持向量机(SVM)模型中用于快速拟合重要的交流(AC)系数,它能够在基本保持原有精度的前提下,有效地提高SVM的回归速度,减少输出参数;提出了直流(DC)系数的按位差分预测编码方法和压缩性能更好的码流产生方法。实验结果表明,该方法不仅缩减了编码时间,而且在相同码率下能获得更好的图像质量。以往基于SV的编码未提及码流结构的描述方法,而快速SV拟合和数据组织方法克服了以往算法可行性差的缺点。
In this paper, an encoding algorithm based on optimal SV （support vector） fitting was presented for low bit-rate image description. The main contributions include： 1） a sample set shrinking strategy was suggested for fast simulation of the most significant AC coefficients in SVM （support vector machine） model. With almost the same accuracy as that of the traditional SVM approximation, the shrinking operation improved the regression speed and reduced the numbers of output parameter. 2） a BBB-DP（bit-by-bit differential prediction） method was suggested for DCs coding, the generation method of optimal bit stream was also proposed. Experiments demonstrated that our method saved coding time, while can acquire images of improved quality images at the same bit-rate. Unlike the previous SV coding, this rapid fitting based method is more feasible and describes the bit-stream structure in detail.
出处
《中国图象图形学报》
CSCD
北大核心
2010年第4期547-556,共10页
Journal of Image and Graphics
基金
国家自然科学基金(60972133)
广东省自然科学基金团队项目(9351064101000003)
广东省能源技术重点实验室项目(2008A060301002)
关键词
极低码率图像压缩
支持向量回归
样本集缩减
按位差分预测
数据流描述
low bit-rate compression, support vector regression （SVR）, sample set shrinking, bit-by-bit differentiate prediction（BBB-DP） , data stream description
