俞洪波
摘 要:在當(dāng)前的信息化社會中,數(shù)字媒體已成為主要的信息載體,并正在滲透到經(jīng)濟發(fā)展、國家安全、社會穩(wěn)定和人民生活的眾多方面,“數(shù)字媒體內(nèi)容平臺”是《國家中長期科學(xué)和技術(shù)發(fā)展規(guī)劃綱要》的優(yōu)先主題。當(dāng)前隨著數(shù)字媒體應(yīng)用的廣泛化和深入化,數(shù)字媒體理解面臨著媒體對象復(fù)雜性、媒體數(shù)據(jù)規(guī)模化、應(yīng)用需求多樣化等挑戰(zhàn)問題,已成為制約數(shù)字媒體應(yīng)用發(fā)展的瓶頸。為了解決這些難題,必須研究媒體內(nèi)容的有效表示、建立符合人類媒體認知的計算模型、充分利用計算機處理的優(yōu)勢,并且實現(xiàn)三者的有機結(jié)合。為達到上述研究目標,需要重點解決3個關(guān)鍵科學(xué)問題:針對媒體認知具有的層次性、整體性,構(gòu)建符合媒體理解層次性和整體性的理論框架;針對媒體對象固有的多義性、多態(tài)性,發(fā)展刻畫媒體對象多義性和多態(tài)性的表示體系;針對媒體計算應(yīng)有的協(xié)同性、高效性,突破制約媒體處理協(xié)同性和高效性的技術(shù)瓶頸。該研究圍繞該項目的三大科學(xué)問題之一——“媒體認知具有的層次性、整體性──如何符合媒體認知的特點”,開展共性基礎(chǔ)科學(xué)問題研究,從視皮層細胞與網(wǎng)絡(luò)水平的信息加工機理、雙光子成像手段、理論建模、計算機仿真等幾個方面研究視皮層神經(jīng)機制是如何支撐視覺認知層次性與整體性的,建立模擬視皮層神經(jīng)機制的算法模型,研究如何在算法層次實現(xiàn)認知層次性與整體性的理論與方法,建立神經(jīng)科學(xué)與計算科學(xué)在視覺問題上的相互促進關(guān)系,并進行實驗驗證,為提高數(shù)字媒體理解的理論與技術(shù)水平奠定基礎(chǔ)。
關(guān)鍵詞:數(shù)字媒體 視覺認知 層次性
Abstract: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, 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.
Key Words:Vision;Hierarchy;Top-down;Bottom-up
閱讀全文鏈接(需實名注冊):http://www.nstrs.cn/xiangxiBG.aspx?id=50117&flag=1