基于分区选择和Gabor小波的遮挡人脸识别

Occlusion face recognition based on partition selection and gabor wavelet

局部遮挡人脸识别是人脸识别应用中的一个难点问题.由于遮挡部分对人脸识别没有贡献,因此在进行分类时应排除这些部分.为了解决这一问题,提出一种将分区选择和Gabor小波相结合的遮挡人脸识别方法.首先,将图像分为互不相连的子块,根据图像均方根误差来确定人脸图像中的遮挡区域;其次,利用5尺度8方向的Gabor滤波器对未遮挡分区图像提取特征;然后,用余弦相似度作为纹理分类器对提取的特征进行识别分类;最后,将测试图像中未遮挡分区的识别结果进行决策融合,得到最终分类结果、统计识别正确率等评级指标.在包含不同遮挡的数据集中测试算法性能,识别准确率均达到95%以上.

Partial occlusion face recognition is a difficult problem in face recognition applications.Since the occlusion parts have a weak contribution to face recognition,these parts should be excluded when classifying.To solve this problem,an occlusion face recognition method combining partition selection and Gabor wavelet is proposed.Firstly,the image is divided into unconnected sub-blocks and the occlusion area in the face image is determined according to the image root mean square error information.Secondly,Gabor filters with 5 scales and 8 directions are used to extract features from unoccluded partitioned images.Then cosine similarity is used as texture separator to identify and classify the extracted features.Finally,the recognition results of the unoccluded partition of the test image are fused for decision making,and the recognition accuracy and other rating indicators are counted.The performance of the algorithm is tested in the data set containing different occlusions,and the recognition accuracy rates reach more than 95%.