摘要
提出了一种基于图像融合的立体图像质量评价方法。通过对立体图像的左右视图进行图像融合生成一幅彩色图像,融合算法采用主成分分析(PCA),使用归一化互相关(NCC)视差图算法,生成了对应的视差图;对融合图像和视差图分别进行归一化亮度系数和谱能量参数的提取,作为支持向量回归(SVR)的输入数据,在经过充分的训练后对立体图像的质量评分进行预测。在LIVE 3D立体图像数据库上的实验结果表明,提出的算法优于最新的无参考立体图像质量评价方法,与人类的主观评价具有较好的一致性。
A method based on image fusion is proposed. Through the left and right view images of a stereo pair, a pair of color image is generated based on principal component analysis (PCA) image fusion. A normalized cross correla- tion (NCC) algorithm is used to generate the parallax image. The normalized intensity coefficient and spectral energy fea- ture extraction of the fusion image and the parallax image are extracted as features. The stereo image quality score is ob- tained by support vector regression (SVR) training. In live 3D image database of experimental results show that the algo- rithm outperforms the latest no reference stereo image quality assessment methods, and has good consistency with human subjective evaluation.
作者
李苗苗
桑庆兵
LI Miaomiao SANG Qingbing(Jiangnan University, Wuxi 214000, China)
出处
《光学技术》
CAS
CSCD
北大核心
2017年第1期25-32,共8页
Optical Technique
基金
国家自然科学基金(61170120)
关键词
立体图像质量评价
无参考
主成份分析
视差图
归一化亮度系数
支持向量回归
stereoscopic image quality assessment
no reference
principal component analysis(PCA)
parallax image(NCC)
normalized luminance eoefficient(MSCN)
support vector regression(SVR)