CIF:RI:Small:Content-Based Strategies of Image and Video Quality Assessment

摘要对无处不在的多媒体访问的需求不断推动编码算法对图像和视频的基于内容的属性进行编码。例如，通过识别和保留感兴趣的区域，或者通过在解码器处合成纹理，可以在保持视觉质量的同时显著降低带宽要求。这些下一代编码策略必须伴随着下一代质量评估算法，可以处理独特的编码工件。然而，以符合人类感知的方式确定质量仍然是一个巨大的研究挑战。目前的质量评估方法使用固定的分析，而人类的感知适应图像？的内容。为了满足日益增长的带宽，移动性和IP流媒体的需求，有一个迫切需要推动国家的最先进的质量评估对这样一个内容自适应的approach.In这项研究中，研究人员进行了一系列的研究，旨在探讨实用的内容自适应模型的人类视觉质量评估的图像/视频包含退化和增强。第一项研究将收集大量增强和退化图像和视频的主观评级。这项工作将为培训和验证提供地面实况数据。利用这些数据，研究者将：（1）研究新的能处理含增强图像的质量评价方法。(2)研究和模拟人类视觉系统在质量评估过程中使用的多种策略，包括基于内容的神经模型和图像自适应策略选择技术。(3)研究质量与感兴趣区域之间的关系。这项研究将带来更准确和更强大的质量评估方法，并将为考虑人类视觉自适应特性的下一代感知模型奠定基础。

AbstractEmerging demands on ubiquitous multimedia access continue to push coding algorithms to ca-pitalize on content-based properties of images and video. For example, by identifying and pre-serving regions-of-interest, or by synthesizing textures at the decoder, it is possible to dramati-cally reduce bandwidth requirements while preserving visual quality. These next-generation coding strategies must be accompanied by next-generation quality assessment algorithms that can handle the unique coding artifacts. Yet, determining quality in a manner that agrees with human perception remains a grand research challenge. Current quality assessment methods use a fixed analysis, whereas human perception adapts to the image?s content. In order to meet increasing demands on bandwidth, mobility, and IP streaming, there is a critical need to push the state-of-the-art in quality assessment toward such a content-adaptive approach.In this research, the investigator conducts a series of studies designed to examine the utility of content-adaptive models of human vision for quality assessment of images/video containing degradation and enhancement. The first study will collect a large set of subjective ratings for enhanced and degraded images and video. This effort will provide ground-truth data for training and validation. Using these data, the investigator will: (1) Research new methods of quality as-sessment that can deal with images containing enhancement. (2) Research and model the mul-tiple strategies employed by the human visual system during quality assessment, including de-veloping content-based neural models and image-adaptive techniques of strategy selection. (3) Research the relationship between quality and regions-of-interest. This research will lead to more accurate and robust methods of quality assessment, and it will lay the groundwork for next-generation perceptual models that take into account the adaptive nature of human vision.