视觉感兴趣区的提取及其在视频图像质量评估中的应用

Detection of region of interest and its application in video image quality assessment

通过主观眼动跟踪实验和客观Itti模型分别研究了视频图像的感兴趣区域提取问题.通过研究眼动跟踪实验数据与视频图像的时间同步问题和适当选取人眼在观察视频时所能接受的感兴趣区域个数,分别得出了主、客观实验的视频图像感兴趣区权重矩阵.在此基础上,在传统图像质量评估方法峰值信噪比(PSNR)中加入主观眼动跟踪实验和客观Itti模型得出的感兴趣区权重矩阵,分析和比较了2种感兴趣区权重矩阵对PSNR的影响.实验证明,通过参数设置,主观眼动跟踪实验和客观Itti模型提取的感兴趣区权重矩阵对PSNR都有明显的改善,改善后的模型不但保持了传统方法的简易性,同时也提高了其与主观感知的相关性.

The detection of region of interest （ROI） in video content is analyzed by the subjective eye-tracking experiment and Itti＇s objective model. For eye-tracking experiments, the time synchronization problem between video content and the data obtained by the eye-tracking experiments is discussed to obtain the subjective weighting matrix of ROI; for Itti＇s model, the optimization of the number of interesting areas is evaluated to deduce the objective weighting matrix of ROI. These two matrices are integrated into the traditional peak signal-to-noise ratio （PSNR） quality assessment metric, the reliability and improvement are discussed. The experimental results show that, by optimizing parameters, the application of the ROI obtained from both eye-tracking experiments and Itti＇s model improves the correlation between the PSNR and subjective assessment while keeping the simplicity.