以整體與局部人臉特徵為主的人臉素描合成系統

機構典藏 > College of Engineering > Graduate Institute & Department of Computer Science and Information Engineering > Thesis > Item 987654321/105729

Please use this identifier to cite or link to this item: https://tkuir.lib.tku.edu.tw/dspace/handle/987654321/105729

Title:	以整體與局部人臉特徵為主的人臉素描合成系統
Other Titles:	Face sketch synthesis system utilizing both global structure and local detailed texture components
Authors:	詹于賢;Chan, Yu-Hsien
Contributors:	淡江大學資訊工程學系碩士班凃瀞珽
Keywords:	人臉素描合成;回歸方程;馬可夫隨機場;Face Sketch Synthesis;Regression;Markov Random Field
Date:	2015
Issue Date:	2016-01-22 15:03:46 (UTC+8)
Abstract:	本文提出一個以線性回歸方程式(2D DCM)為依據的人臉素描合成系統。人臉相片與人臉素描是不同性質的影像，且因為素描是由畫家來繪畫而成的，所以相片與素描之間除了材質上的不同之外還會有幾何上的差異，所以這樣的問題是困難的。傳統的做法主要有兩種方式:(1)著重於怎樣使用影像處理技術來轉換成素描的筆調，這類的做法不會有強調(誇張)個人五官特性的效果(2)使用馬可夫模型來進行合成。而本研究合成的人臉素描則是會與畫家手繪的圖畫相近，會依據收集來的圖片中不同畫家的風格來強調(誇張)人臉個人的特色。在使用線性回歸方程式來合成人臉素描的過程中，我們需要建立一個訓練樣本對資料庫(人臉相片與人臉素描)，並透過線性回歸方程式學習兩種之間的關係式。與傳統馬可夫模型做法不同的是，馬可夫模型由於是以區塊為主來進行合成，所以會導致合成結果有不連續的情況發生。本篇論文主要透過學習好的轉換式來進行轉換，所以不會有這種問題發生。本篇論文主要先利用線性回歸的概念先合成出一張人臉素描，再透過線性回歸與馬可夫隨機場去合成出正確的局部細節，然後與合成出的素描結合起來得到最後結果。我們會將訓練樣本中成對的人臉相片與人臉素描利用線性迴歸方程式去做訓練，得出照片與素描之間的轉換式；如此一來我們就可以將測試影像的照片轉換成一張素描，但由於是透過轉換式達到的結果所以這張素描是模糊的，並沒有正確的人臉細節，接下來我們會透過線性回歸與馬可夫隨機場去補上合成出正確的人臉細節並補上去原本的合成結果。 This study developed a facial sketch synthesis system based on the two-dimensional direct combined model (2DDCM) and the Markov Random Field (MRF) approaches that employs a large collection of photo/sketch pairwise training samples. The proposed synthesis framework achieves facial sketch generation by addressing the following key issues. First, we directly combine each photo and sketch pairwise sample in a concatenated form in order to completely preserve their relationship. Second, photo and sketch images are formed as two–dimensional matrices instead of vectors in order to preserve the facial geometry. Third, both the vertical and the horizontal facial-geometry features are considered in 2DDCM approach. Finally, the proposed framework integrates both the parametric-based 2DDCM approach with the non-parametric-based Markov Random Field (MRF) approach, where the function of former one is to generate facial sketches with correct facial geometry, and the function of later one is used to generate facial sketches with more detailed textures. Accordingly, as the human drawing process, both the global facial geometry and the local detailed textures are included in the synthesis results. Experiments demonstrate our approach can synthesize high quality reconstructed facial sketches from given unseen photos.
Appears in Collections:	[Graduate Institute & Department of Computer Science and Information Engineering] Thesis

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	161	View/Open

Loading...