The integration of early vision, saliency models, and eye-movement control: Experiments, modeling, and spatial statistics

该项目的总体目标是开发一个综合的眼动指导的数学模型，从两个领域的研究，迄今为止一直被视为相当分开的富有成效的组合：早期空间视觉和眼动的动态控制。我们开始从一个神经上合理的和心理评估的人口模型的早期空间视觉的Wichmann的小组在蒂宾根，和一个动态的建模方法，眼动控制和注意力的部署在波茨坦Engbert的小组追求。在项目资助的头两年半里，我们：（一）进行了广泛的实验，测量视觉场景中的眼球运动和注意力转移。(ii)开发了空间相关函数的新分析，以分析视觉场景中的注视模式。(iii)开发了一个生成的动态模型，能够产生丰富的眼动路径。(iv)开发了必要的数学和统计工具的最大似然参数估计在我们的生成动态模型的眼动控制。(v)基于我们的早期空间视觉群体模型，设计了一个视觉显著性模型。(vi)为一个综合数学模型奠定了基础，该模型既能够解释空间视觉数据，又能够预测人眼运动路径。在我们项目的第二个资助或延续阶段，我们计划首先完成空间视觉和眼动控制的综合数学模型的开发。其次，我们计划对我们的模型进行实验测试，受到其生成能力的启发。

The overarching goal of this project is to develop an integrated mathematical model of eye-movement guidance from the fruitful combination of two areas of research which have hitherto been treated rather separately: early spatial vision and the dynamic control of eye-movements. We started from a neuronally plausible and psychophysically evaluated population model of early spatial vision developed in Wichmann's group in Tübingen, and a dynamic modelling approach to eye-movement control and attention deployment pursued in Engbert's group in Potsdam. In the first two and a half years of funding of the project we have: (i) Performed extensive experiments measuring eye-movements and shifts of attention in visual scenes. (ii) Developed new analyses for spatial correlation functions in order to analyse the pattern of fixations in visual scenes. (iii) Developed a generative dynamic model capable of producing rich eye-movement paths. (iv) Developed the mathematical and statistical tools necessary for maximum-likelihood parameter estimation in our generative dynamic model of eye-movement control. (v) Designed a model of visual saliency based on our early spatial vision population model. (vi) Laid the foundations for an integrated mathematical model capable of both explaining data from spatial vision as well as being able to predict human eye-movement paths. In a second funding or continuation phase of our project we plan, first, to finish the development of integrated mathematical model of spatial vision and eye-movement control. Second, we plan to perform experimental tests of our model, inspired by its generative capabilities.