Feature Extraction of Facial Images Using Self-organizing Maps

Abstract

Much information is received from facial expressions during communication. Therefore, during the interface between humans and machines, for example, it's important to extract the information from facial images in order to obtain the correct understanding. This paper proposes a new method for extracting information from facial images. The method provides a self-clustering of image features, and has procedures to cluster images input using self-organizing maps (SOMs). A SOM is an artificial neural network consisting of an input layer and an output layer. When the brightness values of an image are input into the input layer, the output layer works as a map layer that clusters the features of the image. This ability is obtained using a process known as “self-training.” This paper discusses the application of the suggested method for analyzing actual facial images. Six fundamental facial expressions (i. e., happiness, anger, sadness, disgust, fear and surprise) were enacted and recorded by digital VTR. First, an image sequence was composed from the recorded film footage, and the eyebrows, eyes and mouth were semiautomatically selected for observation from each image. Next, training for self-clustering was performed to automatically classify all the facial expressions. After training, an analysis of each SOM was done to review its classification ability. The SOMs were able to classify various shapes of eyebrows, eyes and mouth. When an image sequence from a change in facial expression was input, it was reflected in a change in output that described the change in facial expression. Our method may therefore be used to recognize facial expressions and emotions.