考虑结构一致性和内容完整性的重定向图像质量评价

Image retargeting quality assessment using content integrity and structural consistency

针对适应不同宽高比显示设备全屏显示的重定向图像的视觉质量问题,提出一种基于像素级结构一致性和内容完整性的重定向图像质量评价模型.该模型首先使用后向配准,建立重定向图像和参考图像之间的像素级对应关系;其次考虑到注意力转移的问题,采用像素级宽高比相似性偏差指数计算重定向图像在显著区域和全局的结构一致性,并利用图像组成变化获取图像内容完整性;最后通过内容完整性和结构一致性的线性组合实现重定向图像质量评价.实验结果表明,提出的模型不仅对重定向图像质量评价的效果好,而且可以在内容完整性和结构一致性的二维评估图中观察不同重定向算法的行为,适用于重定向算法比较、选择和优化.

An image retargeting quality assessment model based on pixel-level structural consistency and content integrity is proposed for the visual quality problem of retargeted images for full-screen display of display devices with different aspect ratios.Firstly,this model employs backward registration to establish the pixel-level correspondence between a retargeted image and its reference image.Secondly,the pixel-level aspect ratio similarity deviation index is utilized to calculate the structural consistency of the retargeted image in the salient region and the global towards the attention shift problem.In addition,the image composition changes are adopted to obtain the image content integrity.Finally,the image retargeting quality assessment is approached by a linear combination of the content integrity and the structural consistency.Experimental results show that the proposed CISC model can not only have a good effect on the quality assessment of retargeted images but also observes the behavior of different redirection algorithms in the two-dimensional evaluation graph of content integrity and structure consistency.Hence,the proposed CISC metric is beneficial for the comparison,selection,and optimization of retargeting algorithms.