基于边缘差异的虚拟视图像质量评价方法

摘要

为了降低多视视频中的传输数据量，可适当减少包含纹理图及其对应深度图的多视数量，而在自由立体视频系统的终端采用基于深度图的绘制技术，生成新的虚拟视点。由于不准确的深度图估计、遮挡及合成算法等原因，合成虚拟视的边缘会出现高频噪声。对于这种特有的失真，用一般的评价方法并不能准确反映人眼的视觉感知。该文在分析失真虚拟视与原始视像素差异基础上，对各差异像素进行分类，并分配权重，提出了一种基于视觉加权的边缘差异质量评价方法，对边缘像素施加较高权重。经多视视频序列实验证明，该文方法相比于其它评价方法，能更好地预测人类视觉对虚拟视图像的主观感知。%In order to reduce the transmission data in the multi-view video, the number of multi-views can be decreased properly which include texture images and depth maps, and the intermediate virtual views are generated based on Depth Image Based Rendering (DIBR) at terminal for rendering on 3D displays. Generally, the view synthesis increases the high frequency components on the edge due to inaccurate estimation of the depth map, occlusion and synthesis algorithms. Conventional 2D image quality metric is difficult to reflect the virtual view distortion. In this paper, quality evaluation method based on the edge difference is proposed. Based on the analysis of the pixel difference between virtual and original views, each difference pixel is classified and assigned a visual weight, and higher weights applied to the edge pixels. Experiments for multi-view video sequences prove that the results of this metric are in accordance with characteristic of human visual system.