Case for Support: A Feasability Study into Visual Attention Models for Exploring Complex Information.

该提案提出了一项可行性研究，探讨视觉注意模型的潜力，以增强视觉残疾用户对复杂结构化信息（如图形结构和网络上呈现的电子内容）的可用访问。视障人士经常使用提供语音输出的屏幕阅读器来访问页面内容。默认情况下，屏幕阅读器只是从左上到右下显示内容。考虑到大多数图形结构和网页的复杂性，这样的默认读数是非常不合适的。屏幕阅读器还允许用户通过使用光标键来指导要阅读的内容，但是如果没有对页面的概述，就不清楚如何直观地指导阅读的焦点。然而，视力正常的读者能够将他们的注意力引导到他们最有兴趣探索的电子内容领域。这可以通过快速浏览呈现的数据来实现。视觉注意模型既可以描述也可以预测屏幕上那些在感知上最重要的区域。诸如颜色和对比度信息等因素会引起观察者的注意。这与基于任务的语义知识相结合，可以忽略什么和考虑什么，为有远见的读者提供了一个快速的起点，可以选择如何探索结构化数据。这种视觉注意模型已被用于有选择地指导虚拟环境的部分，这些部分应该以高质量呈现，并赋予虚拟角色看似人类的手势行为。我们希望确定这些是否可以适用于将复杂的图形结构和网页重新呈现给视障读者的新应用环境。我们将通过两个具体的案例研究来探讨这一假设的可行性。第一个是对3D分子结构的探索，第二个是对内容和媒体网站（如新闻网站）的一类网页的再现。两个案例研究中的第一个将根据任务完成时间和评估所呈现数据的非视觉探索有效性的问卷主观结果进行评估。第二项研究将使用两种方法进行评估；首先，通过与基于曼彻斯特现有语义网研究的网页再现的替代方法进行比较，其次，通过与布里斯托尔的视力正常参与者的眼动追踪数据进行比较。如果这些初步实验证明是成功的，一份完整的合作研究项目提案将被提交给UK-EPSRC，以设计一套适合复杂信息非视觉呈现的模型。

This proposal presents a feasibility study into the potential for models of visual attention to enhance usable access by visually disabled users to complex structured information such as graph structures and electronic content presented on the web. Visually disabled people often use screenreaders that provide spoken output to access a page's content. Screenreaders, by default, simply present content from top-left to bottom-right. Given the complex nature of most graph structures and web pages, such a default reading is highly inappropriate. Screenreaders also allow their users to direct what is to be read by use of the cursor keys, yet without an overview of the page it is unclear how the focus of reading could be intuitively directed.Sighted readers, however, are able to direct their attention to the area of electronic content they are most interested in exploring. This is accomplished by obtaining a quick overview of the data presented through a glance. Models of visual attention can both describe and predict those areas of the screen which are most perceptually important. Factors such as colour and contrast information will draw the attention of observers. This coupled with task-based semantic knowledge of what to ignore and what to consider, provides a sighted reader with a quick starting point to choose how to explore structured data.Such models of visual attention have been used to selectively direct portions of virtual environments that should be rendered in high quality and to endow avatars with seemingly human like gesture behaviour. We wish to determine whether these may be adapted to use in the novel application context of re-presenting complex graph structures and web pages to visually disabled readers? We will explore the feasibility of this hypothesis by carrying out two specific case studies. The first being an exploration of a 3D molecular structure and the second a re-presentation of a class of web pages for content and media web sites (such as news sites). The first of the two case studies will be evaluated in terms of task completion times and subjective results from questionnaires assessing the effectiveness of non-visual exploration of the data presented. The second study will be evaluated using two methods; first through a comparison with an alternative method for re-presentation of web pages based on existing semantic web research at Manchester, and second through a comparison with eye-tracked data from sighted participants at Bristol. If these initial experiments prove successful, a full proposal for a collaborative research project to design an implement a suite of models tailored to non-visual presentation of complex information will be submitted to the UK-EPSRC.