Age Estimation by human face Wrinkle Texture representation based on Fractal Dimension features

[[abstract]]在電腦圖學以及機器視覺快速進步下，人臉影像在各種執法、安全控制和人機交互應用估測年齡日益受關注。本論文運用碎形維特徵表現人臉皺紋紋理進行年齡估測研究。人臉皺紋紋理包含內在因素與外在因素影響，每個人老化速度不盡相同具有挑戰性。 本論文以碎形維度作為年齡估測依據，系統架構分為前處理模組、特徵萃取模組、碎形處理模組與分類識別模組四個模組。首先將資料庫影像輸入到系統透過前處理模組Haar-like人臉偵測技術框出人臉，再藉由人臉器官定位出皺紋區域，進行碎形維度演算法則計算。如Differential Box Counting、Shifting-DBC、Scanning-BC。最後進行分類辨識，分類辨識以支持向量機（Support vector machine, SVM）作分類。 本實驗運用FG-NET and PAL兩個資料庫作為實驗用資料庫，這兩個資料庫記錄了圖片相關的年齡與成長資料，包含了不同的人種、年齡、飾物、背景、人臉旋轉角度、視角以及亮度變化和彩色或黑白照片，我們從中挑選出前額區域無遮擋正臉照片來進行測試，包括1152張人臉圖像與96個人不同年齡階段的圖像，有助於年齡估測測試。實驗結果顯示估測準確度DBC、SDBC、SBC分別為70.07%、81.88%、91.07%。[[abstract]]Derived from rapid advances in computer graphics and machine vision, the human face images in a variety of law enforcement, security, application control and human-computer interaction is receiving increasing attention to estimate age. In this paper, Age Estimation by human face wrinkle texture representation based on fractal dimension features. Face wrinkle texture contains internal and external factors, in which everyone with not the same aging rate is a challenging task. This thesis uses a fractal dimension as the basis of age estimation, system architecture are divided into preprocess module, feature extraction module, fractal process module and classification module. First, the images input are fed into the system through pre-processing module which detects face area via Haar-like method. The wrinkle area is positioned by locating organs on the face and the features are computed by using fractal dimension algorithm. Such as Differential Box Counting（DBC）, Shifting-DBC（SDBC）, and Scanning-BC（SBC）. Then, we use SVM for classification. In this study, two face databases use FG-NET and PAL, are need in the experiments. These, databases contain the pictures related to age and growth data, including different races, ages, accessories, background, face rotation angle, viewing angle and brightness variations and color or black and white photos. We pick out the forehead area unobstructed tested positive face photographs, including images 1152 face images and 96 individuals of different ages to help estimate the age test .The databases do help to test our age estimation system. The experimental results demonstrate that the estimation accuracy using DBC, SDBC and SBC may reach 70.07%, 81.88% and 91.07%, respectively.[[note]]碩