請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/62011
標題: | 使用低解析度人臉影像進行性別、年齡與種族分類 Demographic Classification based on Low-Resolution Facial Images |
作者: | Tsung-Hao Ku 顧宗浩 |
指導教授: | 吳家麟(Ja-Ling Wu) |
關鍵字: | 流行對齊,超解析度,性別分類,年齡分類,種族分類, Manifold Alignment,Super-Resolution,Gender Classification,Age Classification,Race Classification., |
出版年 : | 2013 |
學位: | 碩士 |
摘要: | 近年來,使用人臉影像進行人口特性的分類,性別、年齡與種族,
是在多媒體研究裡一項熱門的主題。它在使用者互動系統,或是市場 策略中都扮演著重要的角色。但是高解析度的人臉影像並非永遠唾手 可得,舉例來說,若人臉太遠離攝影機的鏡頭或是攝影的硬體限制都 會產生低解析度的人臉影像。在目前的文獻中,我們是第一位進行處 理高解析度與低解析度影像的連結,在性別、年齡與種族分類問題。 在這項研究中,我們推出了一項系統,能夠快速地處理此類問題,而 不是直觀地使用超解析度(Super Resolution)去處理低解析度影像。 我們使用了流型對齊(Manifold Alignment)直接連接高與低解析度影 像。在我們設計的實驗中,能夠證明我們的方法不但在準確率上高於 傳統的演算法,而且在速度上也能夠超越。 Image-based demographic classification from human faces has been an active topic of multimedia research over the past few years because of its fundamental role in creating a wide context of applications, such as user-aware interaction and strategic marketing planning. However, facial images of high resolution are not always available. For example, due to the hardware limitations of consumer surveillance cameras or people standing at a distance with their faces in small size, it is possible to cause the so-called Low Resolution (LR) problem. To the best of our knowledge, this work is the first in the literature to study the demographic classification problem with a focus on the connection between high resolution (HR) and LR images. Instead of using Super-Resolution (SR) as a preprocessing step, intuitively, to upsample LR images first, in this work, we developed an efficient framework to identify the demographic information, including age and gender, by employing the Manifold alignment techniques to connect the LR and the HR image spaces directly. In the experiments, the proposed approach is evaluated on a public dataset FERET and compared with the baseline algorithm. The results showed that our “one-step” framework can not only achieve better classification performances but also better time efficiency, which implies the proposed approach is more suitable for practical usage. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/62011 |
全文授權: | 有償授權 |
顯示於系所單位: | 資訊工程學系 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-102-1.pdf 目前未授權公開取用 | 4.57 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。