Please use this identifier to cite or link to this item: https://ah.nccu.edu.tw/handle/140.119/102345


Title: 彩色影像之人臉角度分類
Other Titles: Pose Classification of Human Faces in Color Images
Authors: 林群雄
Lin, Chiunhsiun
Keywords: 人臉偵測;人臉角度分類;皮膚顏色分割;權值面具函數
Face Detection;Face Poses Classification;Skin Color Segmentation;Weighting Mask Function
Date: 2001-12
Issue Date: 2016-09-29 17:02:17 (UTC+8)
Abstract: 在本文中,我們建議的系統是由兩個主要部分組成。第一個部分是藉由皮膚顏色分割和等腰三角形為基礎來搜尋潛在臉的區域。我們首先讀取一張RGB彩色影像。先判斷此RGB彩色影像是否有複雜背景。若無,則我們將跳過「皮膚顏色分割方法」,而直接將原RGB彩色影像直接轉變成二值化的影像。若有複雜背景,則將藉由皮膚顏色分割,找出皮膚顏色區域。再將這個皮膚顏色分割後的影像轉變成二值化的影像。再藉由尋找等腰三角形的關係去得到潛在臉的區域。第二部分是要完成臉部角度分類的任務。我們首先將每一個潛在臉的區域,都做了尺寸標準化的處理。然後,藉由人臉權值面具函數獲得每一個人臉的正確位置。其次,再藉由方向權值面具函數判斷人臉的正確方向。最後,再藉由角度權值面具函數決定人臉轉的角度。實驗結果顯示約百分之九十九的成功比率,並且相對錯誤比率很低。
In this paper, we introduce a novel approach for automatic estimation of the poses/degrees of human faces embedded in complicated environments. The proposed system consists of two primary parts. The first part is to search the potential face regions. First, if the input image contains complex background, then the potential face regions are gotten from skin-color- segmentation and the isosceles-triangle criterion that is based on the rules of "the combination of two eyes and one mouth". If the input image contains complex background, then we will use the input RGB color image to perform the human-skin color-segmentation task to remove the complicated surroundings. Then the result of the input image that is removed the complicated surroundings will be converted to a binary image. If the input image doesn't contain complex backgrounds, then we will skip the human-skin color- segmentation task. The input image will be directly converted to a binary image. Secondly, label all 4-connected components and detect any 3 centers of 3 different blocks that form an isosceles triangle. Then, clip the regions that satisfy the isosceles triangle criteria as the potential face regions. The second part of the proposed system is to perform the task of pose verification. In the second part, each face candidate obtained from the previous process is normalized to a standard size (60*60 pixels). Then, each of these normalized potential face regions is fed to the face weighting mask function to obtain the location of the face region. Next, the face region is fed to the direction weighting mask function to determine which direction the matching face region looks at. Last, the face region is fed to the pose weighting mask function to decide the poses/degrees of the human faces. The proposed face poses/degrees classification system can determine the poses of multiple faces embedded in complicated backgrounds. Experimental results demonstrate that an approximately 99% success rate is achieved and the relative false estimation rate is very low.
Relation: 國立政治大學學報, 83, 197-222
Data Type: article
Appears in Collections:[第83期] 期刊論文

Files in This Item:

File Description SizeFormat
fb160923122822.pdf1761KbAdobe PDF505View/Open


All items in 學術集成 are protected by copyright, with all rights reserved.


社群 sharing