模拟生物视觉机制的彩色人脸识别方法

为了充分利用彩色图像提供的信息提高人脸识别的性能,提出了一种模拟生物视觉机制的彩色人脸识别方法。该方法首先构造一种模拟人类的色彩感知机制的对立色模型,将彩色人脸图像描述为对立色形式。然后,模拟初级视皮层的信息处理机制,从图像对立色描述的亮度分量和色度分量分别提取人脸的纹理特征和色彩特征。最后,分别对纹理特征和色彩特征进行分类识别,并将二者的识别相似度融合得到最终的人脸识别结果。该方法利用对立色模型提高了色彩特征对光照变化的鲁棒性,并且综合利用彩色图像的色彩和纹理信息提高了人脸识别的精度,特别是对模糊图像的识别精度。在彩色FERET人脸库和AR人脸库上的实验表明,相对于直接对灰度图像进行识别的方法,该方法对清晰图像的识别率提高了4.5%～16.3%,而对模糊图像的识别率提升更加显著。

Color face recognition method inspired by biological visual mechanisms

In order to improve the performance of face recognition,a color face recognition method inspired by biological visual mechanisms is proposed,which makes extensive use of the information provided in the color image.This method devises an opponent color model,which emulates the color perception mechanism of human,to represent the color face image in an opponent color form.Then,a set of facial texture feature and a set of facial color feature are separately extracted from the intensity component and the chromatic component of the image opponent color representation using a way that simulates the information processing mechanism of the primary visual cortex.The texture feature and color feature are classified respectively,and their similarity scores are fused to obtain the final result of face recognition.The proposed method improves the color feature robustness to illumination variation by means of the opponent color model,and boosts the recognition accuracy,especially the accuracy when the image is blurred,through simultaneously using the texture information and color information of the image.Experiments on the color FERET and AR face databases show that compared with the method that directly classifies gray images,the proposed method improves the recognition accuracy by 4.5%~16.3% when the images are clear,and boosts the accuracy even more when the images are blurred.