针对表情对人脸识别准确率的影响问题,提出了基于向导的局部表情弱化模型,其利用网格变形技术来减少表情人脸的塑性变形.首先在表情区域按三角面片变换,采用梯度算子计算出变形后梯度场,实现基于向导微分梯度场的变换；然后代入泊松方程完成离散三角形拼接和变形模型的重建.找到脸部表情不变的刚性区域,得到类内的平均人脸差异,从而产生约束条件并将之加入变形过程来保持类间的差异和类内的相似度.与其它算法比较,其识别准确率有明显提高,证明了算法的有效性.%For expression impact on face recognition accuracy, a guide-based model was proposed to weaken local expression, and the mesh deformation technology was used to reduce the plastic deformation of facial expression. First the expression area was transformed according to triangular patches,and the deformation gradient was calculated according to gradient operator to realize the guide-based differential gradient field transform, and then the Poisson equation was used to complete the triangle stitching and reconstruction. The rigid region in expression face was found to get the class average face differences and constraints. The constraints was used to maintain the differences between classes and within-class similarity. Comparing with other algorithms, the face recognition accuracy rate is improved significantly,and the effectiveness is proved.
展开▼
