A study on the description model of facial expression and its application to model based coding

研究了基于等密度图的面部表情描述和分类方法。到目前为止，几乎所有的面部表情识别方法都依赖于面部元素的形状和边缘（眉毛、眼睛、鼻子、嘴巴等的位置）。然而，这些特征点的提取和用特征点来描述皱纹等密度变化的变化是非常困难的。与这些方法相比，本研究采用并定义了能够表示面部图像密度变化的等密度图。在该方法中，等密度映射被描述为树形结构。由于等密度图的变化表现为树结构的变化，因此可以通过分析树结构的变化来检测面部表情的变化，例如通过Brother Splitting、Brother merged、Father-Son Splitting或Father-Son merged。该方法的过程如下：1)从人脸图像中提取等密度图2)从等密度图中表示树状结构3)根据树状结构的变化检测人脸表情变化研究了一种提取树状结构变化的描述和比较方法，并通过实验验证了该方法的有效性。从实验结果可以看出，人脸表情的特征变化是可以被提取出来的。这些变化已经在微笑和愤怒的脸上发现，因为皱纹等密度变化变化很大。然而，很难发现惊讶和悲伤的面孔的特征变化，因为它们对皱纹的影响很小。由于本文提出的树匹配是一种基于等密度线位置的方法，在代表小等密度线的节点之间很难找到对应的方法。下一步，必须开发能够检测面部表情微小变化的树匹配算法。

Methods of desciption and classification of facial expression based on isodensity maps are investigated. Until now, almost all methods of recognition of facial expression have depended on the shapes and edges on facial elements (position of eyebrows, eyes, nose, mouth, etc.). However it is very difficult to extract such feature points and to describe the change of density variations such as wrinkles by using feature points. In contrast to these methods, this study adopts and defines isodensity maps that can represent density variation of facial images.In this method, isodensity map is described as a tree structure. As isodensity map changes appear as changes in the tree structure, facial expression changes can be detected by analyzing the changes in the tree structure for example, by Brother Splitting, Brother Merging, Father-Son Splitting, or Father-Son Merging.The processes in this method are as follows :1) Extraction of an isodensity map from a facial image2) Representaion of the tree structure from an isodensity map3) Detection of facial expression shanges based on the changes in the tree structureA description and comparison method for extracting structure changes of tree is investigated and its efficacy is demonstrated experimentally. From experimental results, it is clear that the characteristic changes of facial expression can be extracted. These changes have been found in smiling and angry faces, because of large change in the density variation such as wrinkles. It is, however, difficult to find the characteristic change about surprise and sad faces which have little influence on wrinkles. As the tree matching proposed in this study is a method based on the position of isodensity line, it is difficult to find corresponding method between nodes representing small isodensity lines.At the next stage, tree matching algorithm which can detect small change of facial expression must be developed.