位平面分解的人脸识别集成方法研究

Ensemble Methods for Face Recognition Based on Bit-plane Decomposition

研究了基于集成技术的人脸识别,主要包括集成个体分类器的生成与分类结果的融合.为了提高分类器个体间的差异性,通过位平面分解及移动窗口技术生成基分类器,然后对所分解模式的各层进行融合,以确定最后的决策,这些融合方法包括乘法规则、和规则、多数投票规则、最大值规则、最小值规则与中值规则.另外,针对ORL数据库,实验研究了不同融合方法的人脸识别的性能,并对不同的人脸识别方法的性能进行了比较.结果表明,在基于位平面分解的人脸识别集成方法中,应用和规则的融合方法,其性能也优于其他的人脸识别方法.

Face recognition is studied based on ensemble technique which mainly consists of constructing individual classifiers and fusing classifical results.In order to improve diversity between individual classifiers,the idea of bit-plane decomposition is used and Moving Window Classifier（MWC） is considered as a basic individual classifier.Then the decomposed pattern layers are fused jointly to make a final decision.The used fusion methods include product rule,sum rule,majority vote rule,max rule,min rule and median rule. Moreover, experimental studies are conducted with face images in ORL databases to show face recognition＇ s per- formance based on ensemble learning. And performance of other different face recognition＇ s methods is com- pared with one of face recognition method based on ensemble technique. Experimental results show that performance of using sum rule fused method for face recognition based on bit-plane decomposition is supe or to one of other face recognition＇ s method.