侯春萍, 林洪湖
AuthorsHTML:侯春萍, 林洪湖
AuthorsListE:Hou Chunping, Lin Honghu
AuthorsHTMLE:Hou Chunping, Lin Honghu
Unit:天津大学电气自动化与信息工程学院，天津 300072
Unit_EngLish:School of Electrical and Information Engineering, Tianjin University, Tianjin 300072, China
Abstract_Chinese:针对立体图像失真会改变图像对比度、边缘和结构等信息的特性, 提出一种无参考立体图像质量评价方法．首先, 利用韦泊分布的尺度参数β和形状参数η分别描述合成图、左右视图及视差图的梯度图像特征; 然后用网格强度及其规律性分别描述合成图、左右视图及视差图的网格图像特征; 最后, 将所得特征输入支持向量回归(SVR)网络中训练模型, 获得特征到质量分数之间映射关系的模型, 从而预测立体图像质量．通过利用LIVE-3D I数据库和LIVE-3D Ⅱ数据库进行性能测试, 证明了本文提出的方法与人眼视觉特性具有很高的一致性, 且性能优于目前主流的图像质量评价算法．
Abstract_English:Stereoscopic image distortion will change the contrast，edge，structure and other information of image. To solve the problem，a no-reference stereoscopic image quality estimator was proposed. First，the gradient image characteristics of the cyclopean image，left and right view and disparity map were parameterized using β(scale parameter)and η(shape parameter)of Weibull distribution. Second，the grid characteristics of image were described by grid strength and regularity. Finally，the model described the relationship between the acquired features and scores was trained by support vector regression(SVR)network，which was used to predict the stereoscopic image quality. Experimental results on LIVE-3D I and LIVE-3D Ⅱ show that the proposed method is highly correlated with the human visual system and superior to the mainstream state-of-the-art IQA metrics.
Keyword_Chinese:立体图像质量评价; 韦泊分布; 网格特征; 支持向量回归
Keywords_English:stereoscopic image quality assessment; Weibull distribution; grid characteristic; support vector regression

PDF全文下载地址:http://xbzrb.tju.edu.cn/#/digest?ArticleID=6090