“數字媒體理解的理論與方法研究”項目立項報告

李波

摘 要：當前隨著數字媒體應用的廣泛化和深入化，數字媒體理解面臨著媒體對象復雜性、媒體數據規模化、應用需求多樣化等挑戰問題，已成為制約數字媒體應用發展的瓶頸。為了解決這些難題，必須研究媒體內容的有效表示、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勢，并且實現三者的有機結合。為達到上述研究目標，需要重點解決3個關鍵科學問題：針對媒體認知具有的層次性、整體性，構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、多態性，發展刻畫媒體對象多義性和多態性的表示體系；針對媒體計算應有的協同性、高效性，突破制約媒體處理協同性和高效性的技術瓶頸。圍繞這3個關鍵科學問題的解決，該項目將研究內容分為科學問題解決、關鍵技術攻關、典型應用示范等3個層次，認知機理與計算模型、表示框架與特征描述、融合機制與學習算法、驗證平臺與應用示范等4個方面，設置了以下6個課題：（1）視覺認知的層次性與整體性機制；（2）媒體認知的層次化計算理論與模型；（3）面向多義性對象的學習理論和方法；（4）多模態高維異構數據的特征提取與描述方法；（5）跨媒體分析的理論和方法；（6）數字媒體理解驗證平臺與應用示范。

關鍵詞：數字媒體理解 層次化計算模型 跨媒體分析

Abstract：With the development of digital media technology， how to understand the contents of the media data complexity， large-scaled and application demands diversification has become a very serious problem which restricting the development of the digital media application. In order to solve these problems， the combination of the effectively presentation of media content； establishing the human cognitive media mode and the way to make full use of computer processing advantages is required. In order to solve the problem in a better way，we need to focus on three key scientific issues： how to construct and recognize the hierarchical integrity frame for media technology；how to shape and develop the polysemy、polymorphism frame for media technology； how to break through the media processing restrictions efficiently and cooperatively.To solve these three key issues， the project was divided into three different levels which are： Scientific problem solving， Key technology research and Typical application demonstration； and four different aspects which are： Knowing the principle and calculation model， Frame and feature description， Fusion principle and algorithm learning， Platform verification and application demonstration. We also set up the following six topics：（1）hierarchical and integrity of the visual cognition；（2）computation model and theory for the hierarchical media；（3）the method of multi-modal analysis；（4）description and extraction of the multi-modal high-dimensional data；（5），the method of multi-media analysis；（6）digital media platform verification and application demonstration. This six different topics help us built up the hierarchical model which reveals the hierarchical and integrated characteristics of the media technology； also reveals the inherent law of the polysemy formation of the media technology； building up the multi-dimensional and multi-modal mechanism for the heterogeneous data. Give a new idea of the global and local feature extraction， contextual feature fusion， multi-granularity mapping of the low-level features and high-level semantics， hierarchical semantic analysis of the multi-media； also proposed a number of widely used， smart and high speed processing algorithms. Set up the global high standard media algorithm testing platform， demonstrate the secure and stable network intelligent video surveillance interactive television application to benefit of the public.

Key Words：Digital Media Understanding；Hierarchicalmodel；Across-media Analysis

閱讀全文鏈接（需實名注冊）：http：//www.nstrs.cn/xiangxiBG.aspx？id=50961&flag=1