摘要
为了实现对不同失真类型立体图像的质量评价,提出了一种基于双目能量响应的无参考立体图像质量评价(NR-IAQ)方法。首先,通过对各失真图像进行Gabor滤波,提取出不同频率、不同方向、不同视差响应下的局部特征矢量,作为立体图像特征信息;然后,利用支持向量回归(SVR)建立立体图像特征与主观评价值的关系,从而预测得到立体图像质量的客观评价值。实验结果表明,对于NBU-3D测试库,Pearson线性相关系数值在0.92以上,Spearman等级相关系数值在0.93以上;对于LIVE-3D测试库,Pearson线性相关系数值在0.96以上,Spearman等级相关系数值在0.96以上;与现有的全参考(FR)和(NR)质量评价方法相比,本方法得到的客观评价值与主观评价结果有较好的相关性,更加符合人眼视觉系统。
No-reference image quality assessment is an important task in image processing related applications. In order to achieve quality evaluation for the distorted stereoscopic images with different types of distortion,a no-reference stereoscopic image quality assessment method is proposed based on binocular energy response. In the method,local features are first obtained by applying Gabor filtering on different frequencies,directions and disparities for the distorted stereoscopic images. Then,these local features are fused as stereoscopic features. Finally,support vector regression （SVR） is performed to predict the objective scores by establishing the relationship between the stereoscopic features and the subjective scores. Experimental results show that compared with full-reference and no-reference methods, the Pear- son linear correlation coefficient （PLCC） index reaches 0.92, and the Spearman rank order correlation coefficient （SROCC） index reaches 0.93 for NBU-3D database while the PLCC index reaches 0.96 ,and the SROCC index reaches 0. 96 for LIVE-3D database, respectively, which indicate that the proposed method can achieve higher consistency with subjective assessment of stereoscopic images.
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2014年第4期741-747,共7页
Journal of Optoelectronics·Laser
基金
国家自然科学基金(61271021)
宁波市自然科学基金(2012A610039)
2013年宁波大学研究生科研创新基金(G13053)资助项目
