In human communication, facial expressions play an important role, which carries enough information about human emotions. Last two decades, it becomes a very active research area in pattern recognition and computer vision. In this type of recognition, there is a drawback of how to extract the features because of its dynamic nature of facial structures, which are extracted from the facial images and to predict the level of difficulties in the extraction of the facial expressions. In this research, an efficient approach for emotion or facial expression analysis based on dual-tree M-band wavelet transform (DTMBWT) and Gaussian mixture model (GMM) is presented. Different facial expressions are represented by DTMBWT at various decomposition levels from one to six. From the representations, DTMBWT energy and entropy features are extracted as features for the corresponding facial expression. These features are analyzed for the recognition using GMM classifier by varying the number of Gaussians used. Japanese female facial expression database which contains seven facial expressions; happy, sad, angry, fear, neutral, surprise and disgust are employed for the evaluation. Results show that the framework provides 98.14% accuracy using fourth-level decomposition, which is considerably high.

在人类交流中，面部表情扮演着重要角色，它承载着有关人类情感的足够信息。近二十年来，它已成为模式识别和计算机视觉领域非常活跃的研究领域。在这种类型的识别中，由于其从面部图像中提取的面部结构的动态性质而存在特征提取以及预测面部表情提取中的困难程度的缺点。在这项研究中，提出了一种基于双树M带小波变换（DTMBWT）和高斯混合模型（GMM）的情感或面部表情分析的有效方法。DTMBWT以从1到6的各种分解级别表示不同的面部表情。从陈述中，提取DTMBWT能量和熵特征作为相应面部表情的特征。使用GMM分类器，通过改变使用的高斯数量，分析这些特征以进行识别。日本女性面部表情数据库，其中包含七个面部表情；评估时使用了快乐，悲伤，愤怒，恐惧，中立，惊奇和厌恶。结果表明，该框架使用第四级分解可提供98.14％的准确度，这是相当高的。