基于视觉注意机制的人脸区域预检测

Face Region Pre-detection Based on Visual Attention Mechanisms

为了快速搜索图像中的人脸区域,依据人类视觉系统的特点,引入Itti模型,提出了一种基于肤色特征和方向特征的人脸区域预检测注意力模型,并应用于复杂背景下彩色图像的人脸区域预检测。首先把图像从RGB空间转换到HSI空间,利用肤色在色度空间的聚散性,采用Gaussian金字塔和center-surround算子获取肤色和方向特征的多尺度视觉差异,通过这2个不同特征图的规格化和线性融合获得综合的显著图,由显著图地址转换成图像坐标找出原图中的人脸区域。实验结果表明,该方法能有效检测出人脸区域,且检测速度快,为后续的处理和分析提供了良好的基础。

In order to quickly search the face region in the image, a computing model of face pre-detection is proposed which is based on the characteristics of human face and features skin color and direction. This model would be applied to regional predetection on human face in the color images under complex background. Firstly, the input of the RGB image is converted to the HSI space. Then, the multi-scale features of the visual difference between shin color feature and direction is got by using Gaussian pyramid and the center-surround algorithm, and a comprehensive integration of a significant figure is obtained through this two different characteristics and linear graph normalized. The experimental results show that the method can effectively detect the face region, which laid a sound foundation for subsequent processing and analysis of the face recognition.