CAJ | 학술논문

　　在面部表情识别中，由于图像特征中存在与情感语义无关的信息及噪声干扰等因素，在一定程度上影响表情识别的准确性。传统的基于核典型相关分析的识别方法难以有效克服这些因素的影响。为尽可能排除这些影响表情识别的因素，提出一种基于稀疏表示的核典型相关分析方法，并将其应用于表情识别中。该方法的基本思想是应用稀疏学习方法来自动选择表情特征矩阵中的关键特征谱成分进行表情特征与情感语义特征之间的相关性建模，然后通过建立的模型完成对待测表情图像的语义特征估计，并用于表情的分类识别。为验证所提方法较传统的基于核典型相关分析方法的优越性，选取国际标准表情数据库JAFFE进行实验，实验结果证实了所提方法的有效性。
　　재면부표정식별중，유우도상특정중존재여정감어의무관적신식급조성간우등인소，재일정정도상영향표정식별적준학성。전통적기우핵전형상관분석적식별방법난이유효극복저사인소적영향。위진가능배제저사영향표정식별적인소，제출일충기우희소표시적핵전형상관분석방법，병장기응용우표정식별중。해방법적기본사상시응용희소학습방법래자동선택표정특정구진중적관건특정보성분진행표정특정여정감어의특정지간적상관성건모，연후통과건립적모형완성대대측표정도상적어의특정고계，병용우표정적분류식별。위험증소제방법교전통적기우핵전형상관분석방법적우월성，선취국제표준표정수거고JAFFE진행실험，실험결과증실료소제방법적유효성。
In facial expression recognition, the existences of image noises and the irrelevant image information to the expression changes usually influence the recognition accuracy. The traditional facial expression recognition method using kernel canonical correlation analysis (KCCA) is difficulty to solve this problem. To overcome this drawback, a kernel canonical correlation analysis with sparse representation (SKCCA) is proposed and applied to the facial expression recognition. The basic idea of the SKCCA method is to utilize the sparse representation approach to choose the spectral components of the facial feature matrix before modeling the correlation between facial feature matrix and the expression semantic feature matrix. Then, the expression recognition is carried out based on the correlation model. To demonstrate the superiority of the proposed method over the traditional KCCA method, extensive experiments are conducted on the JAFFE database and the experimental results confirm the effectiveness of the proposed method.