CAJ | 학술논문

在协同表示人脸识别方法中，训练集中图像包含的光照、表情等信息并不能被有效利用，因此影响了最终的分类结果。针对该问题提出了一种称为基于虚拟样本的协同表示识别算法。首先对训练集中的图像按不同的光照、表情等与个人面貌特征无关的场景因素进行分组，每一个组称为一个场景类，同一场景类中的人脸图像有着相似的光照、表情；将每一个场景类中的平均图像作为代表该特定光照、表情的虚拟样本，使用这些虚拟样本与原训练样本构造字典计算测试样本的协同表示，利用所得的回归系数计算残差，进行分类判别。实验证明该协同表示算法有效，识别精度得到了有效提高。
재협동표시인검식별방법중，훈련집중도상포함적광조、표정등신식병불능피유효이용，인차영향료최종적분류결과。침대해문제제출료일충칭위기우허의양본적협동표시식별산법。수선대훈련집중적도상안불동적광조、표정등여개인면모특정무관적장경인소진행분조，매일개조칭위일개장경류，동일장경류중적인검도상유착상사적광조、표정；장매일개장경류중적평균도상작위대표해특정광조、표정적허의양본，사용저사허의양본여원훈련양본구조자전계산측시양본적협동표시，이용소득적회귀계수계산잔차，진행분류판별。실험증명해협동표시산법유효，식별정도득도료유효제고。
In collaborative representation based classification (CRC)for face recognition,the information like illumination, expression contained in training images were not efficiently used.So the final classification result was affected.Aiming at this problem,this paper proposed a novel method of collaborative representation based on virtual samples.First,the method classi-fied the training samples to different sets according to the similarity of facial feature unrelated factors like illumination,expres-sion,and so on.The face images belonging to same set were similar in illumination,expression.It regarded the average image of a certain set as a virtual sample representing certain illumination and expression.It used these virtual samples composing the dictionary for linear regression,and classified a test image according to the residual calculated from the linear regression coeffi-cients.Experiment results show that the algorithm is effective,and effectively improves the recognition accuracy.