基于视觉显著性和视觉信息处理模型的特征提取方法

Face feature extraction based on saliency map and the model of visual information processing

特征提取是模式识别中的一个重要环节,本文首先利用Itti视觉显著性分析算法,提取视觉对输入图像的感兴趣部分,从而一定程度上消除了复杂背景等因素的影响;然后通过视觉信息处理模型HMAX模型对该感兴趣图像通过分层处理提取特征向量,得到了具有尺度不变性、平移不变性、旋转不变性的特征向量,解决了模式识别、尤其是人脸识别中的一个重要难题。在Caltech101中faces库的实验结果证明了该算法的有效性。

Face feature extraction is the key procedure in pattern recognition. First, this paper analyses Itti＇s saliency map for the attention area, which eliminates the influence of environment. Then we get the feature through HMAX model, which is invariant in size, translation and rotation. It solves a difficulty in face recognition. Finally, the experiments on Caltech101 face databases verify the effectiveness of the proposed method.