利用類神經網路於臉部表情辨識

Abstract

本論文提出一種以類神經分類器來自動辨識臉部表情的系統。首先，我們 使用粗略輪廓預測程式 (rough contour estimation routine)}、數學形 態學 (mathematical morphology) 以及點輪廓偵測法 (point contour detection method)這些影像處理的技術，來擷取眉毛、 眼睛和嘴巴這三 個特徵器官的正確輪廓。我們再定義 30 個臉部特徵點 (facial characteristic points)來描述這三個特徵器官的位置和形狀，並產生運 動單元 (action units) 來描述人臉基本的肌肉運動，所以臉部表情可以 由這些運動單元的組合來表示。我們選取六個主要的運動單元當做三種不 同以類神經網路為基礎的表情分類器之輸入向量，而這六個運動單元是由 臉部特徵點的變化所組合而成，所用之表情分類器分別為放射狀基礎函數 網路 (radial basis function network)、回傳網路 (back-propagation network) 和概念模糊集合 (conceptual fuzzy sets)。放射狀基礎函數 網路和回傳網路這兩方法我們可以得到很高的辨識率 92.2\%，用概念模 糊集合則得到了較低的辨識率 60.5%。經由電腦模擬的結果證明了我們所 提出方法的效果。 This thesis proposes an automatic facial expression recognition system usinga neural network classifier. First, we use rough contour estimation routine, mathematical morphology, and point contour detection method to extract the precise contours of the eyebrows, eyes, and mouth of a face image. Then we define 30 facial characteristic points to describe the position and shape of these three organs. Because facial expressions can be described by combining different action units, which are used for describing the basic muscle movement of a human face, we choose six main action units, being composed of facial characteristic points movements, as the input vectors for three different neural network-based expression classifiers including radial basis function network, back-propagation network, and conceptual fuzzy sets. Using radial basis function network and back-propagation network, we have obtained the same recognition rate as high as 92.2\%. Using conceptual fuzzy sets, we have obtained lower recognition rate of 60.5\%. Simulation results by computers demonstrate that computers are capable of extracting high-level or abstract information like human.