skip to main content

運用群眾委外開發表情辨識系統及基準資料

李子杰; Li, Zi-jie 林信志; none; 數位學習科技學系碩士班 2013

線上取得

  • 題名:
    運用群眾委外開發表情辨識系統及基準資料
  • 著者: 李子杰; Li, Zi-jie
  • 林信志; none; 數位學習科技學系碩士班
  • 主題: 群眾委外; 表情辨識; 表情基準資料; 支持向量機; Crowdsourcing; Facial Expression Recognition; Benchmark; SVM
  • 描述: 群眾委外透過特定平台,將需要大量人力參與、但電腦難以取代的重複性任務,委託給網路上一群不特定的自願者;相較於請專人執行任務,群眾委外可以利用人類擅長的能力,提升任務品質與效率,並大幅降低成本。本研究提出一套雙系統機制,可提升表情自動辨識系統的正確率,並發展出一套可運用於其他表情辨識系統的基準資料;此雙系統包括 (1) 群眾委外的社群辨識系統、(2) 電腦運算的自動辨識系統,社群辨識系統利用人類擅長的表情辨識能力,透過群眾委外,讓自願者協助分類臉部影像的表情;為吸引更多自願者參與,本研究將表情辨識任務發展成一款數位遊戲,並以社群辨識結果作為訓練自動辨識系統所需的高辨識度影像;接著,本研究利用 Luxand FaceSDK 擷取每張高辨識度影像的表情特徵,轉換成一個九個維度的向量,再將所有的表情特徵向量輸入支持向量機 (Support Vector Machine, SVM),並分類成高興、悲傷、驚訝、生氣等四個類別;正確歸類的高辨識度影像集,即是最終的表情基準資料。實驗結果顯示:群眾委外的辨識結果可以提升自動辨識到極高的正確率;而具有高辨識度的表情影像集,可以作為其他表情辨識系統的基準資料。
    In this study, a dual system has been proposed to enhance the precision rate of an automatic expression recognition system and, in the meanwhile, to develop a benchmark that can be used to train and test other expression recognition systems. The proposed dual system consists of two systems, including a social expression recognition system using crowdsourcing and an automatic expression recognition system by computer programs. By a well-developed digital game, the social recognition system calls for a lot of online volunteers to classify facial pictures into four kinds of expressions, includes happiness, sadness, surprise, and anger, and thus can collect expression pictures of high validity. Afterward, Luxand FaceSDK is used to extract a 9-dimensional feature vector from each expression picture. These feature vectors are then used to train the SVM classifier in our automatic expression recognition system. The experimental results show that the set of expression pictures collected by crowdsourcing have high validity and can be used to effectively train the automatic expression recognition system to achieve a high precision rate. Also, the set of expression pictures can be used as a benchmark for other expression recognition systems.
    碩士
  • 建立日期: 2013
  • 格式: 121 bytes
    text/html
  • 語言: 中文
  • 識別號: http://nutnr.lib.nutn.edu.tw/handle/987654321/635
  • 資源來源: NUTN IR

正在檢索遠程資料庫,請稍等