基于改进局部五值模式的人脸识别

FACE RECOGNITION BASED ON IMPROVING ELONGATED QUINARY PATTERN

针对局部五值模式EQP(Elongated Quinary Pattern)采用全局阈值定义造成对图像灰度变化敏感以及在人脸识别中对图像不同分块同等对待问题,提出基于局部五值模式增强方法。首先,通过自适应方法来设置阈值,以提高其对图像灰度变化的鲁棒性;其次,通过特征块加权处理,融入每个分块结构对比信息,以突出不同分块的不同作用。采用在人脸识别领域广泛应用的ORL与YALE人脸库进行比较实验,实验结果表明,新方法明显提高了EQP算子的识别效果。

Elongated quinary pattern （EQP） adopts global threshold definition, which causes it being sensitive to the changes in image gray scale. On the other hand, it equally treats different sub-blocks of image in face reeognition. Aiming at these problems, we proposed an EQP-based enhancement method. First, through adaptive approach we determined the threshold to improve its robustness on image gray scale changes. Then, through the treatment of feature blocks weighting we fused the structural contrast information of every sub-block in order to emphasise the different roles of different sub-blocks. We conducted the comparison experiments with the widely used ORL and YALE face database in face recognition field, and experimental results demonstrated that the proposed method greatly improved the recognition effect of EQP operator.