数字媒体理解的理论与方法研究/视觉认知的层次性与整体性机制进展报告

The Theory and Method to Understand the Digital Media/Hierachical and Integrated Mechanism in Visual Cognition

在当前的信息化社会中,数字媒体已成为主要的信息载体,并正在渗透到经济发展、国家安全、社会稳定和人民生活的众多方面,"数字媒体内容平台"是《国家中长期科学和技术发展规划纲要》的优先主题。当前随着数字媒体应用的广泛化和深入化,数字媒体理解面临着媒体对象复杂性、媒体数据规模化、应用需求多样化等挑战问题,已成为制约数字媒体应用发展的瓶颈。为了解决这些难题,必须研究媒体内容的有效表示、建立符合人类媒体认知的计算模型、充分利用计算机处理的优势,并且实现三者的有机结合。为达到上述研究目标,需要重点解决3个关键科学问题:针对媒体认知具有的层次性、整体性,构建符合媒体理解层次性和整体性的理论框架;针对媒体对象固有的多义性、多态性,发展刻画媒体对象多义性和多态性的表示体系;针对媒体计算应有的协同性、高效性,突破制约媒体处理协同性和高效性的技术瓶颈。该研究围绕该项目的三大科学问题之一——"媒体认知具有的层次性、整体性──如何符合媒体认知的特点",开展共性基础科学问题研究,从视皮层细胞与网络水平的信息加工机理、双光子成像手段、理论建模、计算机仿真等几个方面研究视皮层神经机制是如何支撑视觉认知层次性与整体性的,建立模拟视皮层神经机制的算法模型,研究如何在算法层次实现认知层次性与整体性的理论与方法,建立神经科学与计算科学在视觉问题上的相互促进关系,并进行实验验证,为提高数字媒体理解的理论与技术水平奠定基础。

Digital media is the major information source in current world, and it influences the economy, national security and the welfare of the people. Thus it is one of the key issue in 973 projects. When more and more digital media is provided,＆amp;nbsp;the understanding and utilization of these information become a huge challenge, as well as a bottle neck, considering it complexity, data size, and diversified needs. It is urgent to establish the computational model based on human cognition, to set up an effective representation, and to highlight the speed of computer, and integrate them together. It is highly demanding to investigate the following scientific questions.（1）To establish a theoretical framework to reflect the hierarchy and integration of the visual system.（2）To establish a representation system to reflect the diversity of the objects.（3）To break the technical bottle neck for more effective computation. We aimed to investigate the first scientific question, by using multiple techniques, including two photon imaging, computational model, simulation. In detail, we studies the information processing from retina to visual cortex, how the top-down feedback and bottom-up feedforward projections work as a whole, how a computational model simulates this process, and how this computation improves the feature representation. Importantly, the experimental and simulation studies will be compared side by side, to speed up this well-integrated project.