以適應性的主動外形模式定位臉部特徵; Face Feature Locating using Adaptive Active Shape Model

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/9453

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/9453

Title:	以適應性的主動外形模式定位臉部特徵;Face Feature Locating using Adaptive Active Shape Model
Authors:	呂正偉;Cheng-wei Lu
Contributors:	資訊工程研究所
Keywords:	臉部特徵;臉部偵測;主動外型模式;平均模型;face detection;face feature;active shape model;mean shape
Date:	2008-06-26
Issue Date:	2009-09-22 11:48:05 (UTC+8)
Publisher:	國立中央大學圖書館
Abstract:	在近代發表的許多臉部特徵定位方法中，主動外型模式 (active shape model, ASM) 被認為是一個有效的臉部特徵定位工具。在本論文中，我們提供了一個 “以適應性的主動外型模式定位臉部特徵” 的系統。我們的系統包含兩個部份，第一部份是模式訓練，第二部份是定位應用。模式訓練最主要的目的是要從訓練的影像中求出平均模型 (mean shape) 及轉換矩陣 (transform matrix)。而在定位應用中則反覆的執行兩個步驟：(i) 檢查已經找到的特徵點附近的區塊尋找更好的特徵點位置；(ii) 更新模型參數 (shape parameter) 讓模型更符合新找的特徵點位置。我們在定位應用的過程中使用了十字形特徵輪廓 (crisscross profile) 來決定更好的特徵點位置。另外還使用適應性仿射變換函數 (adaptive affine transform) 來得到更好的模型參考位置。在實驗中，我們測試了三種影響偵測效果及效能的因素來比較系統的效能：(i) 特徵輪廓 (profile)、(ii) 固有值的數量 (numbers of eigenvalues)、及(iii) 仿射變換函數 (affine transform)。從實驗結果中，我們可以看到我們所提的系統，在定位臉部特徵的表現上比傳統主動外型模式更出色 Recently, many face feature locating methods have been proposed. Active shape model has been shown to be a powerful tool to aid the interpretation of images, especially in face alignment. In this study, we propose a face facture location system using adaptive active shape model. The proposed system consists of two parts: (1) training process and (2) testing process. In the training process, we train a mean shape and transform matrix from training images. Then the testing process works by alternating the following steps: (i) Examine a region of image around each point for a better position. (ii) Update the shape parameters to fit the new found positions. In order to locate a better position for each point, we utilize the information of crisscross profiles around each point to decide the best position. We also utilize an adaptive affine transform to get a better reference position during testing process. In the experiments, the proposed approaches are evaluated by several different factors such as profiles, numbers of eigenvalues, and two kinds of affine transform. From the experiment results, we find that the proposed approaches can efficiently locate face feature and have better effect than the classical active shape models.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Size	Format

社群 sharing

Loading...