System Flowchart for cloning human facial feature by example

Facial component detection

Skin color detection

Face alignment

Image cloning

Cloning Human Facial Feature By Example 〔基於範例影像之人臉特徵仿製〕


Human face contains a wealth of information, including individual's identity, age, health condition, and emotion. During the last few decades with the vigorous development of computer vision and machine learning techniques, facial feature simulation, face detection and face recognition have been widely applied in real world.
In this thesis, we propose a system on cloning human facial feature based on example images. Our goal is to clone the facial feature of an example image to source image without changing the identity of source image, making source and example image have the same visual effect. In our system, Users first provide the source and example image, and then we use Active Shape Model to find the face contour and facial component shape and exclude them from example, keeping only the facial skin area. Afterward, we adopt skin color detection to remove non-skin color area from facial skin area, and retain only facial skin texture area. Finally, we apply seamless cloning to blend the facial feature of example image to source image, making two images have the same facial feature and achieve the similar visual effect.

SUMMARY (中文總結):

人臉蘊含著豐富的資訊,它可以表達出一個人的身分、年齡、身體狀況、情感。在過去幾十年來,隨著計算機視覺及機器學習等技術的快速發展, 人臉辨識、人臉特徵模擬的技術,已經被廣泛的應用於真實世界中。
在本篇論文中,提出一套基於範例影像之人臉特徵仿製的系統,希望在不改變主體影像的臉部結構下,將範例影像的臉部特徵,轉移至主體影像中, 使主體影像能夠達到與範例影像相似的視覺效果。在此系統中,我們會先讓使用者提供兩張影像,一張為主體影像,另一張為範例影像。之後, 我們採用主動形狀模型 (active shape model) 判斷出臉部及五官的輪廓,並將五官排除,只保留臉部皮膚的區域,再採用膚色偵測 (skin color detection) 從皮膚的區域中,排除非膚色的區域。最後,採用無縫隙仿製 (seamless cloning) 的技術,將範例影像中臉部皮膚紋理的區域合成至主體影像上, 使主體影像與範例影像具有相同的臉部特徵,並達到相似的視覺效果。

PROJECT MATERIAL: (picture gallery, video, software demo, talk slides, etc.)

    Cloning human facial feature from source iamge to example image
    Source Image: Example Image : Result Image (I)
    Source Image: Example Image : Result Image (II)
    Source Image: Example Image : Result Image (III)


    [1] Style Transfer for Headshot Portraits,2014,Data,URL

    [2] drug addicts, URL