基于选择性注意和部分连接神经网络的人脸识别

为了克服以往人脸识别方法因特征提取带来的信息损失与不确定性因素,提出了一种应用于复杂场景中人脸识别方法,这种方法不需要进行特征提取.先对整幅图像使用选择性注意方法,在得到的显著区域中利用Adaboost算法进行人脸搜索与定位,最后将可能包含人脸区域的所有像素全部输入训练好的部分连接神经网络(Parcone)模型进行识别.整个识别过程全部自动完成,不需人工干预,也不必对图像进行预处理.通过利用MIT-CBCL人脸数据库和自建图像库进行的仿真实验表明,该人脸识别方法在复杂背景中具有较高的识别率,可适用于其他类型的目标识别.

Face Recognition Based on Selective Attention and Parcone Model

In order to overcome the loss of information and uncertainties in the previous methods of face recognition, this paper proposes a face recognition method in complex scene, and it does not need feature extraction. The method first used selective attention in the whole image, and then used Adaboost to search and locate the faces in the salience region. Finally, it put all pixels of the region which may contain face into trained partially connected neural evolutionary (Parcone) module to recognize. All of the recognition process was automatically and there is no need for image preprocessing. The experiments use MIT-CBCL face database and self-build image database, and the results show that this face recognition method has good recognition rate in complex background. The method in this paper can be applied to other types of target recognition.