高级检索

基于结构失真分析的立体图像质量客观评价模型

Objective Stereoscopic Image Quality Assessment Model Based on Structure Distortion Analysis

  • 摘要: 在立体图像/视频系统编码、传输、解码等环节中,立体图像质量评价是图像失真的一个重要评判准则.考虑到图像结构特性能较好地反映立体图像质量变化,提出一种基于结构失真的立体图像质量客观评价模型,该模型包括左右视点图像质量和深度感知质量评价两部分.首先根据人眼对图像中不同区域的敏感度存在差异且自然图像具有一定方向奇异性的特性,得到左右视点图像质量评价值;然后利用左右视点的绝对差值图像,采用奇异值向量差异和均值偏差率来描述立体图像深度感的畸变;最后融合前面两部分的评价值得到最终的立体图像质量评价值.实验结果表明,文中模型与主观感知存在较好的一致性,对于高斯模糊、高斯白噪声、JPEG压缩、JP2000压缩以及H.264编码,其评价结果与主观感知之间的相关系数高于0.93,Spearman秩相关系数高于0.92,均方误差低于6.6.

     

    Abstract: Stereoscopic image quality assessment(SIQA) is an important criterion for evaluating distortions in the coding and transmission of stereoscopic image/video systems.Considering the change of image structure information reflects the distortion of stereoscopic images,an objective SIQA(OSIQA) model based on the structure distortion analysis is proposed in this paper.OSIQA combines left-right image quality assessment(LRIQA) with depth perception quality assessment(DPQA).Based on the facts that different regions of an image have different sensitivities,and natural images have direction singularities,the qualities of the left and right views are assessed and further linear weighted to evaluate structure distortion of each view.Moreover,distortion in depth perception is measured with singular value vector and mean error rate with respect to the difference map of the left and right views.Finally,the above two assessing results are combined to describe the quality of the stereoscopic image.The experimental results show that the correlation coefficient(CC) values between the proposed OSIQA model and subjective perception are higher than 0.93 in five distortion types,including Gaussian blur,Gaussian white noise,JPEG,JP2000,and H.264 encoding.Spearman rank order correlation coefficients(SROCC) are higher than 0.92,rooted mean squared error(RMSE) values are lower than 6.6.It means that the proposed OSIQA model is highly consistent with the subjective assessment.

     

/

返回文章
返回