基于思维进化的机器学习的遮挡人脸识别

Face occlusion recognition based on MEBML

针对遮挡人脸的识别问题,提出了基于思维进化的机器学习(MEBML)与局部特征结合的方法。首先提出了LBP偏移特征组的提取方法,定义一种新的特征对比规则,根据遮挡人脸图像与无遮挡人脸图像的局部特征进行对比,记录对比相似度作为局部区域的得分。然后对所有局部区域进行趋同过程和异化过程的演化,得到无遮挡区域及遮挡物体区域。当无遮挡区域满足一定比例且分布集中时,应用该区域特征完成遮挡人脸条件下的人脸识别。实验结果表明:新方法的遮挡人脸识别准确率在92%以上,并具有较低的误识率。

To solve the problem of face occlusion recognition, an algorithm, which combines Mind- Evolution Based Machine Learning （MEBML） and local face feature, is proposed. First, a new method of LBP offset feature group extraction is presented to define a new feature contrast rule. By contrasting images between face occlusion and non-occlusion, the local area score of similarity is obtained. Second, similartaxis and dissimilation evolutionary operations are carried out for all local areas. Then, the areas of face occlusion and non-occlusion are obtained. Finally, if the non-occlusion area is large enough and centralized, the feature of this area can be used to recognize face even it is partially covered. Experiment results demonstrate that the new method can recognize 92% face occlusion images with low error ratio.