媒體認知的層次化計算理論與模型

王蘊紅 薛建茹 黃華 蘭旭光 張兆翔

摘 要：該研究在2013年圍繞“課題計劃任務書”中制定的年度研究內容全面開展工作，重點探討了基于層次性和整體性表達的圖像描述子，基于時空層次化表達的視頻分析，面向整體感知的多層次特征融合機制和基于圖理論的物體表征和匹配。在此基礎上，考慮到媒體認知模型應該對實際視覺問題求解具有指導作用，該課題組本年度面向人臉檢測、行為分析、靜脈和步態識別等實際問題，對相關理論進行了應用示范，探討所研究理論在不同的實際應用場景中的效果，并搭建了一系列驗證平臺予以驗證。主要技術路線有：（1）基于層次性和整體性表達的圖像描述方法；（2）基于時空層次化表達的視頻分析；（3）基于面向整體感知的多層次特征融合機制；（4）基于圖理論的物體表征和匹配幾項。

關鍵詞：媒體認知 層次性 物體識別 視頻分析 圖像匹配

Abstract：The issue surrounding the annual research“project plan mission statement” in the development of fully operational in 2013， focuses on the level and integrity based on the expression of image descriptors， based on video analysis of spatial and temporal hierarchical expression for the overall perception of multi-level feature fusion mechanism and the object representation and matching based on graph theory. On this basis， taking into account the cognitive model of visual problem solving should have a guiding role for our group， we discuss the application of our theory on face detection， behavioral analysis， veins and gait recognition etc. And experimental results show the promising results. The main contents for this report can be explained in 4 parts：（1）Image description based on hierarchy and integrity；（2）video analysis based on hierarchical spatial and temporal expression；（3）multi-level feature fusion mechanism based on overall perception；（4）object presentation and matching based on graph .

Key Words：Hierarchy；Object Recognition；Video Analysis；Image Matching

閱讀全文鏈接（需實名注冊）：http：//www.nstrs.cn/xiangxiBG.aspx？id=50918&flag=1